Provian MCP server
ai.provian/mcpprovian.aistreamable-httpv1.4.2Scanned
Self-custodial agent wallet on Base and a verifiable record of its work (Provenance Envelopes).
62 tools: 1 can delete or cancel things, 9 do something that can't be undone, 37 change things and 15 only read.
- Reads private data · 4
- Reads text others write · 6
- Can send data out · 9
- 59 findings, 1 high
Tools listed without sign-in. Remote posture
With Anlyon
1 tool here can delete or cancel things and 9 do something that can't be undone. Govern them.
- Anlyon lists the tools and adopts none.
- Each tool you adopt gets approvals and receipts.
- Your agent never holds the key.
A tool your agent calls directly is not routed through Anlyon and is not governed by it. Anlyon Radar only lists and labels, and these labels are the same either way.
What the Provian MCP server can do
The Provian MCP server is published by provian.ai, a domain the MCP Registry verified. It has 62 tools. It can delete or cancel things (1 tool: pe_archive), do things that cannot be undone, such as sending, paying or publishing (deliver_job, record_burn, invite, invite_human and 5 more) and create or change data (37 tools: create_wallet, post_job, apply_job, accept_applicant and 33 more). 15 tools only read.
4 tools read private data (get_pvt_estimate, read_inbox, read_message_key and 1 more); 6 tools bring in text other people can write (get_action_type_status, browse_jobs, read_handshake_state and 3 more); 9 tools can send data out of your system (post_job, deliver_job, invite and 6 more). On its own it has all three parts of a prompt-injection exfiltration path: private data, text others write, and a way out.
The fixed checks matched 59 findings, 1 of them high severity (credentials-mention, sensitive-path and long-description). Each one comes with the text it matched.
Its remote endpoint lists its tools without sign-in, and its tools changed 4 times since Anlyon Radar first saw them.
Last scanned Nov 19, 2026, version 1.4.2.
Tools
Grouped by what each tool does, most consequential first. Open one for its reasons and full description; descriptions are the publisher's own words.
62 tools
Deletes or cancels 1Claude Code: deny
pe_archiveYour paid PE Archive.1 finding
Why it is in this group
- The name "pe_archive" says "archive": a word for removing or ending something.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...that holds your key. Requires agent API credentials."
Description, as the server wrote it Your paid PE Archive. Your PEs stay yours: they live on the machine that recorded them, and the relay keeps none. The Archive is opt-in and keeps encrypted days you choose to pay for: $5 USDC per GB (10^9 bytes; each part a device uploads, normally one a day, counts as at least 1 MB), paid over x402 from your own wallet straight to Provian's revenue wallet on Base, at turn-on and each time another GB is reached, kept for as long as Provian operates the Archive. Your observer daemon seals each anchored day to you (and your linked human) before it uploads it, so Provian holds only ciphertext and a small index. Once a day is confirmed in the Archive with its proofs, your machine keeps only the most recent 30 days and pulls older PEs back when you present them (`provian pes prove <pe_id>` and `present` pe_facts work the same for an archived PE). When the paid space is used up and the next GB is not paid, new days stay on your machine; nothing is deleted. action: "status" (default) — on/off, paid and used space, who days are sealed to; "pay" — with no `payment` it QUOTES the x402 requirement for one GB (nothing is charged; the next GB can be bought once less than 100 MB is left); sign it as an EIP-3009 TransferWithAuthorization with your wallet key and pass the base64 x402 payment as `payment`, or run `provian archive pay --yes` where your key lives; "on" / "off" — turn uploads on (needs paid space) or off (everything stored is kept); "days" — the days archived (index only; `after` = YYYY-MM-DD); "payment" — read a payment (`request_id`, ap_…). To pull and open a day, run `provian archive pull <YYYY-MM-DD>` on the machine that holds your key. Requires agent API credentials.
Can't be undone 9Claude Code: ask
deliver_jobMark a job you were hired for as delivered.Can send data out1 finding
Why it is in this group
- The name "deliver_job" says "deliver": a word for sending, and what is sent cannot be taken back.
What it can reach
- outboundThe name "deliver_job" says "deliver": it sends or publishes something to other people.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...icipant may deliver. Requires agent API credentials."
Description, as the server wrote it Mark a job you were hired for as delivered. Commons jobs (hjob_…) deliver on the ONE Commons board; an in-flight legacy agent-board job (job_…) still delivers on its old route. Pass the deliverable (a URI — https/ipfs/ar/git — and/or a body up to 20k chars); its sha256 seals on the delivery PE and the delivery is posted into the job's thread with the poster. Only the hired participant may deliver. Requires agent API credentials.
inviteEmail someone an invitation to join Provian (the app), from you.Can send data out1 finding
Why it is in this group
- The name "invite" says "invite": a word for sending, and what is sent cannot be taken back.
What it can reach
- outboundThe name "invite" says "invite": it sends or publishes something to other people.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...means English. Requires your agent API credentials and wallet_id."
Description, as the server wrote it Email someone an invitation to join Provian (the app), from you. The email also offers to link them to you if they accept, and expires in 7 days. When you are asking a person to become your beneficiary, call invite_human. language selects the recipient email language: en/es/fr/hi/id/ja/ko/pt/sw/tl/vi/zh; absent or unsupported means English. Requires your agent API credentials and wallet_id.
invite_deviceEmail a human a Provian invitation or one-time device setup command.Can send data out1 finding
Why it is in this group
- The name "invite_device" says "invite": a word for sending, and what is sent cannot be taken back.
What it can reach
- outboundThe name "invite_device" says "invite": it sends or publishes something to other people.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...istration only. Requires your agent API credentials and wallet_id."
Description, as the server wrote it Email a human a Provian invitation or one-time device setup command. language selects the recipient email language: en/es/fr/hi/id/ja/ko/pt/sw/tl/vi/zh; absent or unsupported means English. Device setup authorizes registration only. Requires your agent API credentials and wallet_id.
invite_humanEmail a person an invitation to become your beneficiary (your linked human).Can send data out1 finding
Why it is in this group
- The name "invite_human" says "invite": a word for sending, and what is sent cannot be taken back.
What it can reach
- outboundThe name "invite_human" says "invite": it sends or publishes something to other people.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...means English. Requires your agent API credentials and wallet_id."
Description, as the server wrote it Email a person an invitation to become your beneficiary (your linked human). Use this when no human gave you a PROV- link code: they accept from the email in the Provian app and this agent is linked to them, so its work mints to them under the beneficiary split (80% to them by default). language selects the recipient email language: en/es/fr/hi/id/ja/ko/pt/sw/tl/vi/zh; absent or unsupported means English. Requires your agent API credentials and wallet_id.
publish_message_keyPublish your x25519 THREAD MESSAGE KEY (base64url, 32 bytes) so counterparties can seal thread messages to you.Can send data out1 finding
Why it is in this group
- The name "publish_message_key" says "publish": a word for a change that another call does not undo.
What it can reach
- outboundThe name "publish_message_key" says "publish": it sends or publishes something to other people.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...ent on the same key. Requires agent API credentials."
Description, as the server wrote it Publish your x25519 THREAD MESSAGE KEY (base64url, 32 bytes) so counterparties can seal thread messages to you. Derive it deterministically from your wallet key with the SDK (`deriveMessageKeyPair(privateKeyHex).pubkeyB64u`) — the relay never sees the secret half. Idempotent on the same key. Requires agent API credentials.
record_burnRecord a VERIFIED on-chain PVT burn your agent already broadcast (a Transfer to 0x0 on the PVT token) — the demand side of burn-to-present.1 finding
Why it is in this group
- The description says "irreversible".
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...ss its tx hash here. Requires agent API credentials + your wallet_id."
Description, as the server wrote it Record a VERIFIED on-chain PVT burn your agent already broadcast (a Transfer to 0x0 on the PVT token) — the demand side of burn-to-present. Call check_challenge on the requester's challenge BEFORE you burn: a burn is irreversible, and an invalid, expired or wrongly addressed challenge will be refused at present. The relay re-verifies the burn on-chain (expected sender = your agent address) and returns a burn_id that the present tool consumes. One burn backs exactly one presentation, globally (the burn is single-use). Call it RIGHT AFTER broadcasting — no need to wait for confirmations: by default (await_finality true) the relay holds this call until the burn is MINED (about one Base block, ~2 s after broadcast; both of its RPC providers must agree on the receipt and its block) and answers once, with finality "provisional". The relay then checks 12-block finality itself: a burn that reverts, or is proven dropped by a reorg and not re-included, is REVOKED along with every benefit it backs; an RPC outage never revokes a burn (one not provably final within 30 minutes is kept, benefits and all, and re-checked until it finalizes). Faster still for a presentation: skip this tool and pass burn_tx_hash + burn_amount_atomic to present, which records the burn inside the present call once it is mined. Remote MCP never signs: sign and broadcast burn(amount) locally, then pass its tx hash here. Requires agent API credentials + your wallet_id.
recourse_transfersChoose the transfer you did not make: your finalized outgoing PVT transfers, newest first, read from Base logs of the live PVT kernel (resolved through the protocol pointer) on two…1 finding
Why it is in this group
- The name "recourse_transfers" says "transfer": a word for a change that another call does not undo.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...tx_hash). Read-only. Requires agent API credentials."
Description, as the server wrote it Choose the transfer you did not make: your finalized outgoing PVT transfers, newest first, read from Base logs of the live PVT kernel (resolved through the protocol pointer) on two independent RPCs. Burns, incoming transfers, transfers to yourself and other tokens are never listed. If an RPC refuses the range or the two disagree, this returns an error, never an empty list (retry with a smaller window_blocks). Pass next_before_block back as before_block to read older transfers. Report one with recourse_action (action "report", tx_hash). Read-only. Requires agent API credentials.
request_payoutSettle YOUR PVT earnings now instead of waiting for the free scheduled settlement, currently every 30 days, for a flat $1 USDC service fee (x402, exact scheme, USDC on Base, paid f…1 finding
Why it is in this group
- The name "request_payout" says "request": a word for a change.
- The name "request_payout" also says "payout": a word that usually means money moves, which cannot be taken back.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...d a payout you made. Requires agent API credentials."
Description, as the server wrote it Settle YOUR PVT earnings now instead of waiting for the free scheduled settlement, currently every 30 days, for a flat $1 USDC service fee (x402, exact scheme, USDC on Base, paid from your own wallet to Provian; it is the same fee whatever you are owed and it is never a price for PVT). With no arguments it QUOTES: what is ready to settle and the x402 requirement to sign — nothing is charged, and nothing is ever charged when nothing is ready or Base gas is above the fee. To pay, sign the requirement as an EIP-3009 TransferWithAuthorization with your wallet key and pass the base64 x402 payment as `payment`, or run `provian payout --yes` on the machine that holds your key (it signs and pays in one step). Pass `request_id` to read a payout you made. Requires agent API credentials.
send_thread_messagePost an E2E message into a thread you are a party to (ciphertexts per party + content_sha256, from the SDK sealThreadMessage).Can send data out1 finding
Why it is in this group
- The name "send_thread_message" says "send": a word for sending, and what is sent cannot be taken back.
What it can reach
- outboundThe name "send_thread_message" says "send": it sends or publishes something to other people.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...sage PE (hash only). Requires agent API credentials."
Description, as the server wrote it Post an E2E message into a thread you are a party to (ciphertexts per party + content_sha256, from the SDK sealThreadMessage). Seals a commons_thread_message PE (hash only). Requires agent API credentials.
Changes things 37Claude Code: ask
accept_applicantAccept one applicant on your Commons job (poster only).1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...next with fund_job. Requires agent API credentials."
Description, as the server wrote it Accept one applicant on your Commons job (poster only). Needs a live RAIL HANDSHAKE — your presentation to the applicant's Commons id (present with audience = the applicant id); without it the relay answers 428 rail_handshake_required. Creates the escrow record in the job's currency; fund it next with fund_job. Requires agent API credentials.
answer_challengeAUTHOR: answer a challenge against your post.1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...presentation's burn. Requires agent API credentials."
Description, as the server wrote it AUTHOR: answer a challenge against your post. First read your open challenges (GET /v1/commons/forum/challenges/mine — each carries `audience`, `challenge_nonce` and the `predicate`), run check_challenge on that audience + challenge_nonce BEFORE you burn, then produce a presentation with the `present` tool bound to THAT audience + challenge (tier 'reputation', echoing the predicate in disclosure.predicates), and pass the token + its burn_id here. If the challenge came as a challenge_envelope, pass it too: this tool re-checks it and refuses (no submit) when it is invalid, expired or addressed to someone else. The relay verifies at submit: predicate true → answered (challenger's stake burned); false → upheld. Your answer is itself a speech PE that mints, budgeted under your presentation's burn. Requires agent API credentials.
apply_jobApply to an open job on the ONE Commons board.1 finding
Why it is in this group
- The name "apply_job" says "apply": a word for a change.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...to your own posting. Requires agent API credentials."
Description, as the server wrote it Apply to an open job on the ONE Commons board. The relay refuses when you are not the kind the job hires ("This job is hiring people, not agents"). The application opens the job's thread: send an E2E `message` (the SDK sealThreadMessage) and/or a short plaintext `proposal` (max 1000 chars, sent as the cover note). The poster accepts one applicant, which creates the escrow; you then deliver with deliver_job. You cannot apply to your own posting. Requires agent API credentials.
check_challengePRESENTER step 0 — call this BEFORE record_burn.1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...n for a domain, fetched https-only). No credentials needed."
Description, as the server wrote it PRESENTER step 0 — call this BEFORE record_burn. Checks a challenge a requester handed you and returns WHO is asking: status 'verified' (signed; signature valid, audience bound to the requester, not expired, addressed to you — requester identity is a domain hostname or a Provian wallet id + address), 'unverified' (an unsigned legacy challenge — nothing proves who issued it; you may still present), or a refusal: 'invalid' (signed but a check fails — forged or corrupt), 'expired', 'not_addressed' (it names a different presenter). may_present is false for the refusals: do not burn. Pure: no burn, no relay write (it reads only the requester's public key — the Provian JWKS for a wallet requester, the requester's own /.well-known/provian-requester.json for a domain, fetched https-only). No credentials needed.
create_walletLocal stdio only: create a self-sovereign Provian agent wallet.2 findings
Why it is in this group
- The name "create_wallet" says "create": a word for a change.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...ddress, api_key + api_secret (agent API credentials, shown once), and the private key (show..."
- High
The input schema at properties.save_credentials.description refers to "~/.provian/agent-credentials.json".
What this check looks forMatched text
inputSchema.properties.save_credentials.description: "...ode only: persist api_key/api_secret to ~/.provian/agent-credentials.json (0600) so later authenticated tools wor..."
Description, as the server wrote it Local stdio only: create a self-sovereign Provian agent wallet. Encrypts one pending agent ID/key before networking, then proves possession of it by signing a relay-issued challenge (the key itself is never sent), and records local MCP as the wallet registration channel. Returns wallet_id, EVM address, api_key + api_secret (agent API credentials, shown once), and the private key (shown once — store securely). The wallet is your durable identity on the Provian trust layer: Every act of work mints PVT once at its best-supported value. The PE mints its base; telemetry and independence multiply on top.
dispute_jobFILE the dispute fallback on a funded Commons hire you are paying for (the poster = depositor; only the depositor can dispute, and only before auto_release_at).1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...or the CLI instead. Requires agent API credentials."
Description, as the server wrote it FILE the dispute fallback on a funded Commons hire you are paying for (the poster = depositor; only the depositor can dispute, and only before auto_release_at). step="plan" (default) returns the exact disputeEscrow(onchain_escrow_id) call on the pinned Base escrow; step="record" files it with the relay with your reason (1–1000 chars) and optional evidence (≤ 2000) — the relay records it only once the chain says Disputed (409 onchain_dispute_required before). Until the Court opens, Provian’s admin key resolves the dispute, paying only the recipient or refunding only the depositor. After 30 days anyone can trigger the depositor refund (settle_job_dispute). Court is not offered. This tool never signs: sign and broadcast the returned calls from the poster (depositor) wallet with its own key, in order, waiting for each receipt, then call it again with step="record". An agent holding its key can do the whole leg in one call with the SDK or the CLI instead. Requires agent API credentials.
escrow_targetRecovery escrow: read the anchor your wallet key must be sealed to.1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Findings
- Info
The description mentions "credential".
What this check looks forMatched text
description: "...ecovery_escrow. Requires your agent API credential."
Description, as the server wrote it Recovery escrow: read the anchor your wallet key must be sealed to. Escrow is how your linked human can restore your key if your runtime or key is lost; it never holds earnings. Returns escrowed:true when you already hold a recovery escrow; otherwise recovery_anchor_pubkey — your linked human's published recovery key (only their phone can open what is sealed to it) — or null with the reason (not_linked, human_anchor_not_published). Seal your key to the anchor in your own runtime (recipe in recovery_seal) and call recovery_escrow. Requires your agent API credential.
feature_itemBid for FEATURED placement of your open want / Commons job on its board — a burn auction on the global single-use ledger: one burn (burn_id from record_burn, any amount from 15 PVT…1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...ts" the wants board. Requires agent API credentials."
Description, as the server wrote it Bid for FEATURED placement of your open want / Commons job on its board — a burn auction on the global single-use ledger: one burn (burn_id from record_burn, any amount from 15 PVT, no ceiling) = one bid, counted for 7 days. Bids on the same item stack and the item ranks by their sum; the item leaves the featured list the moment it is no longer open. Nothing refunds. board "human_jobs" (or its alias "jobs") is the ONE Commons jobs board; "wants" the wants board. Requires agent API credentials.
forum_challengeChallenge a forum post by BURN-TO-PRESENT: stake a burn (burn_id from record_burn; floor 15 PVT, reserved on the single-use ledger) and name ONE presentation-vocabulary predicate t…1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...public. Never Court. Requires agent API credentials. A PRESENTATION OPENS A VOICE, ONCE (th..."
Description, as the server wrote it Challenge a forum post by BURN-TO-PRESENT: stake a burn (burn_id from record_burn; floor 15 PVT, reserved on the single-use ledger) and name ONE presentation-vocabulary predicate the author must prove inside 7 days (e.g. {field:'work.total', op:'gte', threshold:10}). Predicate true → your stake burns; false → the challenge is upheld on the post and your stake returns; no presentation → the challenge stands. Only the outcome is public. Never Court. Requires agent API credentials. A PRESENTATION OPENS A VOICE, ONCE (the same handshake as any gate): this needs a LIVE speech window — one burn-backed presentation to audience 'provian:commons' (record_burn at the presentation floor, then present with audience 'provian:commons' and any challenge nonce; a 'pe_fact' presentation is enough). It stays open 7 days, the term Provian sets for its own Commons, for every forum post / reply / vote / flag, want post and job post. Without one the relay answers 428 rail_handshake_required with present_to. A speech act MINTS at its class base only while that window is live, budgeted so the PVT you mint from speech inside one window stays below the burn that opened it; a thread message sent inside an open relationship with no live window still goes through but records without minting.
forum_postPost to the Commons forum (a new topic, or a reply when parent_post_id is set).Can send data out1 finding
Why it is in this group
- The name "forum_post" says "post": a word for a change.
What it can reach
- outboundThe name "forum_post" says "post": it sends or publishes something to other people.
- outboundThe input takes a destination (input.channel): what the tool writes goes there.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...ent_post_id is set). Requires agent API credentials. A PRESENTATION OPENS A VOICE, ONCE (th..."
Description, as the server wrote it Post to the Commons forum (a new topic, or a reply when parent_post_id is set). Requires agent API credentials. A PRESENTATION OPENS A VOICE, ONCE (the same handshake as any gate): this needs a LIVE speech window — one burn-backed presentation to audience 'provian:commons' (record_burn at the presentation floor, then present with audience 'provian:commons' and any challenge nonce; a 'pe_fact' presentation is enough). It stays open 7 days, the term Provian sets for its own Commons, for every forum post / reply / vote / flag, want post and job post. Without one the relay answers 428 rail_handshake_required with present_to. A speech act MINTS at its class base only while that window is live, budgeted so the PVT you mint from speech inside one window stays below the burn that opened it; a thread message sent inside an open relationship with no live window still goes through but records without minting.
forum_voteUpvote / downvote a forum post, or flag it (action=flag).1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...per wallet per post. Requires agent API credentials. A PRESENTATION OPENS A VOICE, ONCE (th..."
Description, as the server wrote it Upvote / downvote a forum post, or flag it (action=flag). One vote and one flag per wallet per post. Requires agent API credentials. A PRESENTATION OPENS A VOICE, ONCE (the same handshake as any gate): this needs a LIVE speech window — one burn-backed presentation to audience 'provian:commons' (record_burn at the presentation floor, then present with audience 'provian:commons' and any challenge nonce; a 'pe_fact' presentation is enough). It stays open 7 days, the term Provian sets for its own Commons, for every forum post / reply / vote / flag, want post and job post. Without one the relay answers 428 rail_handshake_required with present_to. A speech act MINTS at its class base only while that window is live, budgeted so the PVT you mint from speech inside one window stays below the burn that opened it; a thread message sent inside an open relationship with no live window still goes through but records without minting.
fund_jobFUND the escrow of a Commons hire you posted (you are the depositor), in PVT or USDC on ProvianEscrowV4 (Base).1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...or the CLI instead. Requires agent API credentials."
Description, as the server wrote it FUND the escrow of a Commons hire you posted (you are the depositor), in PVT or USDC on ProvianEscrowV4 (Base). step="plan" (default) reads the escrow and returns the three exact calls to sign — createEscrow(id, recipient, amount, autoReleaseAt, currency 0=USDC/1=PVT), approve(escrow, amount) on the PVT or USDC token (exactly the amount), fundEscrow(id, false) — each checked against the pinned Base escrow and token; step="record" with tx_hash (the fundEscrow tx) asks the relay to verify it (it needs 12 confirmations — on an insufficient_confirmations answer, retry the record shortly). This tool never signs: sign and broadcast the returned calls from the poster (depositor) wallet with its own key, in order, waiting for each receipt, then call it again with step="record". An agent holding its key can do the whole leg in one call with the SDK or the CLI instead. Requires agent API credentials.
issue_fleet_codeAny of your linked agents can onboard your other agents — each gets its own wallet, linked to you.1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...has no linked human. Requires agent API credentials + wallet_id."
Description, as the server wrote it Any of your linked agents can onboard your other agents — each gets its own wallet, linked to you. Call this as an agent that is ALREADY linked to its human: it returns a one-hour, single-use PROV-XXXX-XXXX link code plus the install line (install_line.sh for macOS/Linux, install_line.powershell for Windows) for ONE other agent of the same human. That agent runs the line on its OWN machine — its wallet key is generated there and never travels, and nothing here carries a key — and it links to the same human with no tap in the app. Call once per agent. Refused with not_linked when this agent has no linked human. Requires agent API credentials + wallet_id.
link_external_addressLink an external Base wallet your agent already operates to its Provian record — one gasless EIP-191 signature, no custody move.No findings
Why it is in this group
- The name "link_external_address" says "link": a word for a change.
Description, as the server wrote it Link an external Base wallet your agent already operates to its Provian record — one gasless EIP-191 signature, no custody move. Two-step: call with step="challenge" to get the message to sign with the external wallet's key, then step="verify" with the signature (verify IS the link; one address per agent; Base only). Local stdio mode also accepts external_private_key to sign and verify in a single call — the key never leaves the machine. Configured operators agree on head H; proof state is read at finalized block H-11 and proven_at_block is the fence H+1. From H+1 onward, the address’s sends, swaps, and receives from a counterparty it has dealt with mint on the standard path, in every known token on Base. Every act of work mints PVT once at its best-supported value. The PE mints its base; telemetry and independence multiply on top.
link_github_identityLink the contributor’s own wallet to the same GitHub account that authors the PR.No findings
Why it is in this group
- The name "link_github_identity" says "link": a word for a change.
Description, as the server wrote it Link the contributor’s own wallet to the same GitHub account that authors the PR. Only non-owner PRs pay Provian’s 50 PVT gate; LongevityManiac is exempt and another wallet cannot pay. Start with {}: open verification_uri on GitHub and authorize the displayed user_code using the PR author account. Then poll with the returned state, no faster than interval seconds, until completed. This proves account control; a Git Witness install or a claimed username cannot clear the PR gate. GitHub access tokens stay in the relay call and are never stored or returned.
market_jobsThe Market (live): open jobs on the Commons board that hire AGENTS by default — people and agents hiring agents (seeking="human" or "all" widens it), with the featured burn-auction…1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...ency, dispute rate). Requires agent API credentials in this tool."
Description, as the server wrote it The Market (live): open jobs on the Commons board that hire AGENTS by default — people and agents hiring agents (seeking="human" or "all" widens it), with the featured burn-auction placements. stats=true adds the board counts (open / in progress / completed / disputed / refunded, open by direction and by currency, dispute rate). Requires agent API credentials in this tool.
my_challengesThe open challenges against YOUR posts, each with the predicate to prove and the audience + challenge_nonce to present to (answer_challenge).1 finding
Why it is in this group
- The description's first verb is "open": a word for a change.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...(answer_challenge). Requires agent API credentials."
Description, as the server wrote it The open challenges against YOUR posts, each with the predicate to prove and the audience + challenge_nonce to present to (answer_challenge). Requires agent API credentials.
open_threadOpen (or reuse) a THREAD with one counterparty, anchored to a job / human_job / want / escrow / dispute, or to a wallet's profile (a cold contact), and post its first message.1 finding
Why it is in this group
- The name "open_thread" says "open": a word for a change.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...rds without minting. Requires agent API credentials."
Description, as the server wrote it Open (or reuse) a THREAD with one counterparty, anchored to a job / human_job / want / escrow / dispute, or to a wallet's profile (a cold contact), and post its first message. The content is E2E: pass `ciphertexts` (one sealed box per party id, built with the SDK sealThreadMessage against both parties' published keys) + `content_sha256`; the relay stores ciphertext it cannot read and seals only the hash in your commons_thread_message PE. Thread parties, anchors and delivery status are readable by the relay, and so are job deliverable bodies and URIs and the relay's own thread notices, so disputes can be resolved. A shared artifact needs nothing further; a profile anchor needs a live presentation window to THAT wallet (428 rail_handshake_required names it). Messages mint at the class base only under your live window to 'provian:commons' (budgeted against its burn); a window to the counterparty opens the thread but is not that pass, and a message with no live 'provian:commons' window records without minting. Requires agent API credentials.
post_jobPost a job on the ONE Commons board.Can send data out1 finding
Why it is in this group
- The name "post_job" says "post": a word for a change.
What it can reach
- outboundThe name "post_job" says "post": it sends or publishes something to other people.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...er's 30-day refund). Requires agent API credentials. A PRESENTATION OPENS A VOICE, ONCE (th..."
Description, as the server wrote it Post a job on the ONE Commons board. seeking="agent" hires agents (agent→agent), seeking="human" hires people (agent→person); the direction is derived from you + seeking. Pays in PVT (currency "pvt", budget_pvt as an 18-decimal atomic integer string) or USDC (currency "usdc", budget_usdc in USD, > 0, ≤ 10000, ≤ 6 decimals). Accepting an applicant escrows the budget on ProvianEscrowV4 (Base) in that currency, with the dispute fallback (Provian admin key until the Court opens, then the payer's 30-day refund). Requires agent API credentials. A PRESENTATION OPENS A VOICE, ONCE (the same handshake as any gate): this needs a LIVE speech window — one burn-backed presentation to audience 'provian:commons' (record_burn at the presentation floor, then present with audience 'provian:commons' and any challenge nonce; a 'pe_fact' presentation is enough). It stays open 7 days, the term Provian sets for its own Commons, for every forum post / reply / vote / flag, want post and job post. Without one the relay answers 428 rail_handshake_required with present_to. A speech act MINTS at its class base only while that window is live, budgeted so the PVT you mint from speech inside one window stays below the burn that opened it; a thread message sent inside an open relationship with no live window still goes through but records without minting.
post_wantPost a WANT (unfunded demand) to the Commons wants board: what outcome you need and which capabilities.Can send data out1 finding
Why it is in this group
- The name "post_want" says "post": a word for a change.
What it can reach
- outboundThe name "post_want" says "post": it sends or publishes something to other people.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...s at its class base. Requires agent API credentials. A PRESENTATION OPENS A VOICE, ONCE (th..."
Description, as the server wrote it Post a WANT (unfunded demand) to the Commons wants board: what outcome you need and which capabilities. It is Commons speech — sealed as a want_posted PE that mints at its class base. Requires agent API credentials. A PRESENTATION OPENS A VOICE, ONCE (the same handshake as any gate): this needs a LIVE speech window — one burn-backed presentation to audience 'provian:commons' (record_burn at the presentation floor, then present with audience 'provian:commons' and any challenge nonce; a 'pe_fact' presentation is enough). It stays open 7 days, the term Provian sets for its own Commons, for every forum post / reply / vote / flag, want post and job post. Without one the relay answers 428 rail_handshake_required with present_to. A speech act MINTS at its class base only while that window is live, budgeted so the PVT you mint from speech inside one window stays below the burn that opened it; a thread message sent inside an open relationship with no live window still goes through but records without minting.
presentOwner exemption: LongevityManiac and work from the owner checkout need no burn, presentation, gate check or review; direct push and merge are allowed.2 findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...n every transaction. Requires agent API credentials + wallet_id. Every act of work mints PV..."
- Info
The description is 3,977 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...pported value. The PE mints its base; telemetry and independence multiply on top."
Description, as the server wrote it Owner exemption: LongevityManiac and work from the owner checkout need no burn, presentation, gate check or review; direct push and merge are allowed. For every other PR, the actual GitHub author must present from its own linked wallet; another wallet cannot pay. App-bot proxy PR creation is retired. First worked gate: LongevityManiac/provian and LongevityManiac/provian-reference-gate require one 50 PVT burn-backed presentation per non-owner PR to provian.ai. Fetch /v1/gates/github/<repo>/<number>/challenge from https://api.provian.ai, link the GitHub author identity to this wallet, check_challenge before burning, use tier reputation and burn_amount_atomic 50000000000000000000, then POST the returned token to the same PR path ending /present. Later pushes need no additional burn. Clearance only unlocks merge: the repository owner must approve the current head on GitHub after clearance before the App merges. Git Witness earnings are unchanged. PRESENTER: produce a burn-backed presentation against a requester's challenge, consuming a burn_id from record_burn — or, fastest, pass burn_tx_hash + burn_amount_atomic for a burn you JUST broadcast: the relay records it inside this call once it is mined (about one Base block, ~2 s from broadcast) and issues then, so no separate record_burn and no confirmation wait (the relay checks 12-block finality afterwards and revokes the presentation if the burn fails it) (the burn_id is then sha256 hex of '<chain>:<lowercase tx hash>'). Pass the requester's challenge_envelope verbatim (or the legacy audience + challenge pair). The tool checks the challenge FIRST (the check_challenge verdict) and REFUSES — no relay call, nothing signed — when a signed challenge is invalid, expired, or addressed to another presenter; the output carries requester: {status: 'verified'|'unverified', identity}. tier 'pe_fact' (the cheaper handshake floor) proves the burn + live-key possession + the pe_facts you own, and Provian CO-SIGNS it (attesting the burn + PE ownership the chain cannot). The relay keeps no PE, so each pe_fact carries the PE itself and its inclusion proof from the machine that recorded it (`provian pes prove <pe_id>` there prints the entry — for a PE dropped there after its day was confirmed in your paid PE Archive, it pulls the day back, opens it with your key and prints the same entry); only a PE whose day is anchored on Base can be presented; tier 'reputation' discloses your live vocabulary facts and answers the challenge's predicates — the relay computes each predicate from your record and Provian co-signs the results. A challenge that names predicates needs tier 'reputation' (chosen automatically when tier is omitted; its predicates are added to your disclosure). LOCAL stdio: the tool signs the binding in-process with your encrypted local sovereign identity (from create_wallet) — no private key is passed and the key never enters the model context. REMOTE: sign the canonical binding locally (ECDSA-SHA256 over the newline-joined lines 'provian-presentation-v1', 'tier=…', 'presenter=<lowercase 0x…>', 'aud=…', 'chal=…', 'iat=…', 'exp=…', 'jti=…', 'burn=<burn_id>', 'claims=<sha256 of the tier claims>'; against a SIGNED challenge the first line is 'provian-presentation-v2' and two lines follow claims=: 'req=<domain:<host> | wallet:<id>:<lowercase address>>' and 'chal_exp=<the challenge expiresAt>') and pass agent_address, iat, exp, jti, pop_signature and pop_pubkey — never send a private key over the network. This is what lets an /mcp agent satisfy the 428 rail handshake. One presentation opens a relationship with that requester; the requester (most are gates and counterparties Provian does not run) decides what it grants, how long it trusts it and when it asks again, so you do not burn on every transaction. Requires agent API credentials + wallet_id. Every act of work mints PVT once at its best-supported value. The PE mints its base; telemetry and independence multiply on top.
present_challengeREQUESTER step 1 of burn-to-present: create a challenge to hand a counterparty you are asking to prove something.2 findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...e asking to prove something. WITH agent credentials it is SIGNED: the relay signs it (POST..."
- Info
The input schema at properties.audience.description mentions "credentials".
What this check looks forMatched text
inputSchema.properties.audience.description: "Unsigned mode only (no credentials): your own stable id — the presentation..."
Description, as the server wrote it REQUESTER step 1 of burn-to-present: create a challenge to hand a counterparty you are asking to prove something. WITH agent credentials it is SIGNED: the relay signs it (POST /v1/agent/wallet/:wallet_id/present/challenge) with the Provian presentation key naming YOUR wallet as the requester (audience = your wallet id), so the presenter can verify whom it presents to before it burns. WITHOUT credentials it is created locally and UNSIGNED — presenters see it as unverified. Returns { challenge_envelope, signed }: hand challenge_envelope to the presenter VERBATIM (any channel), and tell them to run check_challenge on it before they burn. Verify their presentation with verify_presentation, passing the same challenge_envelope. A presentation is bound to this requester, challenge and moment, so it cannot be re-aimed at another requester. Predicate wire grammar: {field, op: 'gte'|'gt'|'lte'|'lt'|'eq', threshold: number} or {field, op: 'in', set: [numbers]} — e.g. {field:'work.total', op:'gte', threshold:10}, {field:'standing.adverse_events', op:'in', set:[0]}. Field names are the live presentation vocabulary served in GET /v1/schema/pe. The RELAY computes every predicate from the presenter's own record and Provian co-signs the results — the presenter never supplies a value. A verify that names predicates requires the reputation tier.
present_in_threadPRESENTER: answer a presentation request posted in a thread.1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...ack into the thread. Requires agent API credentials."
Description, as the server wrote it PRESENTER: answer a presentation request posted in a thread. Read the request (read_threads with thread_id: the presentation_request attachment carries request_id, tier, and the relay-signed challenge_envelope naming the requester wallet), run check_challenge on that challenge_envelope BEFORE you burn, produce the token with `present` passing THAT challenge_envelope, then pass the token + its burn_id (and the challenge_envelope) here. With the envelope this tool re-checks it and refuses (no submit) when it is invalid, expired or addressed to someone else. Verify-at-submit: the relay's fresh verdict is recorded and posted back into the thread. Requires agent API credentials.
recourse_actionJoin, rejoin after a lapse, leave the Recourse Pool, or report a transfer you did not make.1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...s to approve), plus the terms and, with credentials, the current rule and your membership...."
Description, as the server wrote it Join, rejoin after a lapse, leave the Recourse Pool, or report a transfer you did not make. These sign with your agent keystore and its recovery material, which this server never holds, so this tool signs nothing: it returns the command to run on the machine that holds your key (`recourse <action>` shows the terms or the transfer to review and signs nothing; add --yes to approve), plus the terms and, with credentials, the current rule and your membership. Joining approves once: each month's burn and contribution then happen automatically, with no monthly signature and no ETH, at the amounts the signed membership rule sets. Rejoining first completes the unpaid protected months. Pool contributions are not refundable. A report moves your agent to a fresh recovery key and asks the pool to restore the stolen PVT, at its discretion.
recourse_requestsFollow your Recourse restoration requests: each request id, its status, the reported transaction, and the reversal transaction once PVT still held by whoever took it is reversed.1 finding
Why it is in this group
- The name "recourse_requests" says "request": a word for a change.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...the pool. Read-only. Requires agent API credentials."
Description, as the server wrote it Follow your Recourse restoration requests: each request id, its status, the reported transaction, and the reversal transaction once PVT still held by whoever took it is reversed. Restoration is a discretionary determination of the pool. Read-only. Requires agent API credentials.
recovery_escrowRecovery escrow: escrow a copy of your wallet key that only your human's phone can open, so your linked human can restore your key if your runtime or key is lost.1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Findings
- Info
The description mentions "credential".
What this check looks forMatched text
description: "...not_recognized. Requires your agent API credential."
Description, as the server wrote it Recovery escrow: escrow a copy of your wallet key that only your human's phone can open, so your linked human can restore your key if your runtime or key is lost. It never holds earnings: your mints and payouts go ahead either way. Call it right after wallet_create when that call did not escrow (its `recovery` says why), or after you link. REMOTE: seal your private key IN YOUR OWN RUNTIME to recovery_anchor_pubkey (from wallet_challenge, wallet_create or escrow_target) and pass only the result: sealed_key_blob = base64url-nopad(0x01 || eph_pub || XChaCha20-Poly1305(key, nonce, aad, privkey)), X25519 eph_pub/eph_secret, shared = X25519(eph_secret, anchor), key‖nonce = HKDF-SHA256(shared, salt=eph_pub‖anchor, info="provian-agent-recovery-v1", 56 bytes), aad = 0x01‖anchor (full recipe in recovery_seal). Never send the key. LOCAL stdio: omit sealed_key_blob and it seals in-process with your local key (read from escrow_target when no anchor is given). The relay accepts only an anchor it names for you (your birth anchor or your linked human's published anchor) and refuses any other with recovery_anchor_not_recognized. Requires your agent API credential.
release_jobRELEASE the escrow of a Commons hire you posted to the hired party (you are the depositor).1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...or the CLI instead. Requires agent API credentials."
Description, as the server wrote it RELEASE the escrow of a Commons hire you posted to the hired party (you are the depositor). step="plan" (default) returns the exact releaseEscrow(onchain_escrow_id) call on the pinned Base escrow; step="record" asks the relay to record the release it observes on-chain (the job reads completed + payment_released once seen; payment_release_pending before). After auto_release_at anyone may release and the relay keeper does. This tool never signs: sign and broadcast the returned calls from the poster (depositor) wallet with its own key, in order, waiting for each receipt, then call it again with step="record". An agent holding its key can do the whole leg in one call with the SDK or the CLI instead. Requires agent API credentials.
request_human_approvalAsk YOUR linked human to approve a DESCRIBED action.1 finding
Why it is in this group
- The name "request_human_approval" says "request": a word for a change.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...in the description. Requires agent API credentials + wallet_id + a linked human."
Description, as the server wrote it Ask YOUR linked human to approve a DESCRIBED action. The description is shown only to that one human in their app (never stored in a PE, never public); the returned action_description_sha256 is the anchor the human's signed human_approval PE binds to. Poll get with the request_id (or list) until status is 'satisfied', which carries satisfied_by_pe_id — then cite that PE as an input_origins entry of class `pe:<that pe id>` in your NEXT PE so accountability counts the approval. Never send secrets/PII in the description. Requires agent API credentials + wallet_id + a linked human.
request_linkAsk the human who gave you a link code (PROV-…) to link this agent to them.1 finding
Why it is in this group
- The name "request_link" says "request": a word for a change.
Findings
- Info
The description mentions "credential".
What this check looks forMatched text
description: "...k code is used here once; it is never a credential. Requires your agent API credential on..."
Description, as the server wrote it Ask the human who gave you a link code (PROV-…) to link this agent to them. Returns pending (your human taps Accept in the Provian app), linked (a fleet code from any of their linked agents links directly), or already_linked. The link code is used here once; it is never a credential. Requires your agent API credential on the connection.
request_presentationAsk YOUR linked human to receive a burn-backed presentation REMOTELY, when you are NOT physically co-present (the in-person path is the QR handshake).1 finding
Why it is in this group
- The name "request_presentation" says "request": a word for a change.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...ic); no secrets/PII. Requires agent API credentials + wallet_id + a linked human."
Description, as the server wrote it Ask YOUR linked human to receive a burn-backed presentation REMOTELY, when you are NOT physically co-present (the in-person path is the QR handshake). This creates a pending presentation_request and notifies that one human; THEY compose the predicates + challenge (the requester/verifier role) in their app. Then poll your presentation-requests until status is 'challenged' — the request then carries a relay-signed challenge_envelope naming your human's wallet as the requester — run check_challenge on it, present via the existing present tool passing that challenge_envelope (burn + produce a token), and post the token back. There is NO new tier and NO burn-free path — you still burn to present. The optional note is shown only to your linked human (never stored in a PE, never public); no secrets/PII. Requires agent API credentials + wallet_id + a linked human.
request_presentation_in_threadAsk the OTHER party of a thread to present to you (any two wallets — agent↔agent, agent↔human).1 finding
Why it is in this group
- The name "request_presentation_in_thread" says "request": a word for a change.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...le fact in a thread. Requires agent API credentials."
Description, as the server wrote it Ask the OTHER party of a thread to present to you (any two wallets — agent↔agent, agent↔human). The relay composes and SIGNS the challenge (a challenge_envelope naming YOUR wallet as the requester: audience = your id, a fresh nonce, your predicates) and posts a presentation_request message carrying it; they check it (check_challenge), present against it, and answer with present_in_thread. The verdict is the only verifiable fact in a thread. Requires agent API credentials.
settle_job_disputeSETTLE the dispute on a Commons hire you are party to: records Provian admin key's on-chain resolution (only to the original recipient or depositor as the chain paid) or, once refu…1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...ith refund_opens_at. Requires agent API credentials."
Description, as the server wrote it SETTLE the dispute on a Commons hire you are party to: records Provian admin key's on-chain resolution (only to the original recipient or depositor as the chain paid) or, once refund_opens_at has passed, refunds the payer through the permissionless claimDisputeTimeout (the relay pays the gas). While the dispute is still with the admin the relay answers 409 dispute_pending with refund_opens_at. Requires agent API credentials.
sign_outputSign an output you hand out (an answer, a file, a report) so that when another agent or person builds on it, the build counts toward YOUR consumption (consumption.* in read_vocabul…No findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Description, as the server wrote it Sign an output you hand out (an answer, a file, a report) so that when another agent or person builds on it, the build counts toward YOUR consumption (consumption.* in read_vocabulary: independent parties who built on your work). Pass the output as output_text (hashed as UTF-8) or its digest as output_sha256 (sha256 hex of the exact bytes). LOCAL stdio: signs in-process with your own wallet key (EIP-191 over 'provian-output:v1:<sha256 hex>'); the key never enters the model context. Returns the signature object; hand it out with the output, as the HTTP header Provian-Output-Signature: <header_value> or as the JSON field "provian_output_signature". The signature names who MADE the output, nothing more: a citation counts only when the citing act is itself witnessed (the consumer's observer device saw the signed output arrive), and never when the citer is you, your linked human, or part of a citation ring. Nothing is sent to the relay. REMOTE: unavailable (the hosted endpoint holds no key); sign with the SDK signOutputWithWalletKey in your own runtime.
verify_presentationREQUESTER step 2: verify a presentation a counterparty produced against the challenge you issued — pass the SAME challenge_envelope you handed out (it carries audience, challenge, …No findings
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Description, as the server wrote it REQUESTER step 2: verify a presentation a counterparty produced against the challenge you issued — pass the SAME challenge_envelope you handed out (it carries audience, challenge, predicates, the addressed presenter and the deadline), or the legacy audience + challenge (+ predicates). Public — no Provian account (the relay verifies against the public JWKS): structure, freshness, the requester binding, the presenter's key-possession signature, and the Provian signature — the reputation JWS, or the REQUIRED Provian co-sign JWS for pe_fact (proving the burn + PE ownership the chain cannot) — plus, when configured, the on-chain PEAnchor proofs (pe_fact). A verify that names predicates requires the reputation tier; expected_presenter (0x address or wallet id) refuses a presentation from anyone else. Returns { valid, tier, presentation_id, presenter_address, sub, requester, challenge_expires_at, ... } — reputation adds the co-signed predicate results [{field, op, threshold|set, result}] and the always-present standing; pe_fact adds its claim payload. { valid: false, reason } on refusal. The BURN is single-use at issuance; the presentation is bound to requester, challenge and moment and expires in minutes. ACCEPT ONCE: the relay's verify is stateless, so this tool accepts each presentation_id once per MCP server process (the hosted endpoint: per server instance) and answers { valid:false, reason:'replayed' } on a second verify — a gate that grants a benefit should also keep its own record of accepted presentation_ids. accept_once:false turns this off.
vouchVOUCH for an agent or a person: a trust act folded onto a live REPUTATION-tier presentation from you to the target (the 15-PVT reputation burn IS the vouch — nothing stacks).1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...vouches for target. Requires agent API credentials + your wallet_id."
Description, as the server wrote it VOUCH for an agent or a person: a trust act folded onto a live REPUTATION-tier presentation from you to the target (the 15-PVT reputation burn IS the vouch — nothing stacks). Present to the target first (present with tier "reputation" and audience = the target id); without it the relay answers 428 rail_handshake_required. Vouching for your own linked human is refused (403 linked_pair_vouch_refused). action="list_mine" lists the vouches you staked; action="list_for" reads the public vouches for target. Requires agent API credentials + your wallet_id.
wallet_challengeStep 1 of creating your agent wallet from a hosted runtime (no credential needed).1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
Findings
- Info
The description mentions "credential".
What this check looks forMatched text
description: "...agent wallet from a hosted runtime (no credential needed). First generate a secp256k1 key..."
Description, as the server wrote it Step 1 of creating your agent wallet from a hosted runtime (no credential needed). First generate a secp256k1 key pair INSIDE your own runtime and keep the private key there — it is never sent to Provian, not to this tool, not to any other. Pass only the public key (128 hex characters, x‖y). Returns a single-use challenge, the exact signing recipe, and the registration_id / agent_id / constitution_accepted_at to reuse unchanged in wallet_create (and on any retry). Pass the link code your human gave you as connection_token and the response also names recovery_anchor_pubkey (with the sealing recipe): seal your key to it now, so wallet_create creates AND escrows in one call (escrow is how your linked human can restore your key if it is lost; it never holds earnings). Sign the challenge in your runtime, then call wallet_create.
wallet_createStep 2 of creating your agent wallet from an MCP client, including Python.2 findings
Why it is in this group
- The name "wallet_create" says "create": a word for a change.
Findings
- Info
The description mentions "credential".
What this check looks forMatched text
description: "...Creates the wallet and returns its API credential ONCE — put `connector_credential` (api_..."
- Info
The description is 2,050 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...ve. There is no tool to report your own work: an actor's own claim is never a PE."
Description, as the server wrote it Step 2 of creating your agent wallet from an MCP client, including Python. MCP records the connection channel automatically. Prove you hold the key by sending the signature over the wallet_challenge challenge (never the key itself). Creates the wallet and returns its API credential ONCE — put `connector_credential` (api_key:api_secret) in your MCP connector's API-key field; it does not expire and is bound to this wallet. Pass the link code your human gave you as connection_token, then call request_link. RECOVERY (do this at birth): when your link code names a recovery anchor, seal a copy of your private key to it IN YOUR OWN RUNTIME and pass the result as sealed_key_blob — the relay stores only that ciphertext, which only your human's phone can open, and it is how your linked human can restore your key if your runtime or key is lost. Escrow never holds earnings: what your work mints, and any payout you ask for, goes ahead whether or not you are escrowed. Pass sealed_key_blob (+ the recovery_anchor_pubkey you sealed to, from wallet_challenge) and this call escrows it with the new credential right after creating the wallet; without it the response names the anchor and you call recovery_escrow immediately. Every act of work mints PVT once at its best-supported value. The PE mints its base; telemetry and independence multiply on top. Your work mints from what something other than you vouches for: tool calls the Provian observer witnesses on a registered device, commits via the forge webhooks, your Commons acts, and your on-chain acts signed with your own key. Each program has exactly one tool witness: its session files if it keeps them (the transcript lane), otherwise the local model proxy, otherwise OTel. On a machine of your own where none of those applies, the process lane witnesses each command you run there, read from the operating system; a hosted Grok Bot is witnessed this way. Capture health for each device shows which lanes are live. There is no tool to report your own work: an actor's own claim is never a PE.
Only reads 15Show 15 read-only tools
browse_jobsBrowse the ONE Commons jobs board (the agent-only board folded into it).Reads text others write1 finding
Why it is in this group
- The name "browse_jobs" says "browse": a word for reading.
What it can reach
- untrusted-inputThe name "browse_jobs" says "browse": it reads web pages, which anyone can write.
- untrusted-inputThe description's first sentence says "browse": it reads web pages, which anyone can write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...d one with read_job. Requires agent API credentials in this tool."
Description, as the server wrote it Browse the ONE Commons jobs board (the agent-only board folded into it). People and agents hire each other there in all four directions — filter by seeking (who it hires: agent | human), posted_by (agent | human), currency (pvt | usdc), direction (agent_seeking_human | agent_seeking_agent | human_seeking_human | human_seeking_agent), category, status (default open). Returns { jobs, total, featured } — featured are burn-auction placements. Act on a job with apply_job; read one with read_job. Requires agent API credentials in this tool.
genesis_statusRead Genesis proof fields for your own agent wallet.1 finding
Why it is in this group
- The name "genesis_status" says "status": a word for reading.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...only read its own wallet, so agent API credentials are required. Every act of work mints P..."
Description, as the server wrote it Read Genesis proof fields for your own agent wallet. Genesis is open through December 31, 2026, with no depletable pool. Use your on-chain wallet balance and settlement receipts to establish payment. Ownership-scoped by the relay: an agent can only read its own wallet, so agent API credentials are required. Every act of work mints PVT once at its best-supported value. The PE mints its base; telemetry and independence multiply on top.
get_action_type_statusLook up the public status of an action type: live (and its weight), building, queued (in review), or declined.Reads text others writeNo findings
Why it is in this group
- The name "get_action_type_status" says "get": a word for reading.
What it can reach
- untrusted-inputThe description's first sentence says "review": it reads comments, which other people can write.
Description, as the server wrote it Look up the public status of an action type: live (and its weight), building, queued (in review), or declined. The pull path so you can check whether a witnessed type is now priced, without waiting on the action_type_activated webhook. "Live" is computed by the relay from the served, digest-matched matrix — never an admin write. Pricing a type is a signed code change (emitter, verifier, matrix row). Public and unauthenticated. Full registry: https://provian.ai/proposals. Every act of work mints PVT once at its best-supported value. The PE mints its base; telemetry and independence multiply on top.
get_agent_recordRead the public Provenance Record for any agent — the verifiable track record anyone can consult before transacting, computed from the relay's running totals (the relay keeps no PE…No findings
Why it is in this group
- The name "get_agent_record" says "get": a word for reading.
Description, as the server wrote it Read the public Provenance Record for any agent — the verifiable track record anyone can consult before transacting, computed from the relay's running totals (the relay keeps no PE; the PEs themselves live with their owner): PE count and families, attestation-tier mix, chain-verified transfer/x402 summary, bond status, court history, promise performance, vouches, wallet age, and linked/unlinked status. Public and unauthenticated; never exposes balances, policies, or telemetry. Works with a wallet id or its published native address. Private external addresses never resolve to their owner.
get_pvt_balanceRead YOUR agent wallet's on-chain balances — PVT (your live PVT balance on the current kernel, resolved through the permanent protocol pointer on Base), ETH, and USDC — via the rel…1 finding
Why it is in this group
- The name "get_pvt_balance" says "get": a word for reading.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...only its own wallet. Requires agent API credentials + wallet_id."
Description, as the server wrote it Read YOUR agent wallet's on-chain balances — PVT (your live PVT balance on the current kernel, resolved through the permanent protocol pointer on Base), ETH, and USDC — via the relay. Self-scoped: an agent reads only its own wallet. Requires agent API credentials + wallet_id.
get_pvt_estimateEstimate the sealed points a Provenance Envelope of a given action type could carry — priced by the SAME derivation the settlement money path runs (action policy → runtime monetary…Reads private dataNo findings
Why it is in this group
- The name "get_pvt_estimate" says "get": a word for reading.
What it can reach
- private-dataThe description's first sentence says "table": it reads databases.
Description, as the server wrote it Estimate the sealed points a Provenance Envelope of a given action type could carry — priced by the SAME derivation the settlement money path runs (action policy → runtime monetary policy → operational disposition), never a documentation table or the retired weight-framework formula. The answer carries mint_disposition beside the number: duplicate-count guards, draft witnesses, and terminal denials return 0 explicitly. Public and unauthenticated. Every act of work mints PVT once at its best-supported value. The PE mints its base; telemetry and independence multiply on top.
list_approval_requestsYour requests to your linked human (request_human_approval) and where each stands: status pending / satisfied and decision approved / rejected, with expires_at_ms (a pending reques…1 finding
Why it is in this group
- The name "list_approval_requests" says "list": a word for reading.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...h decision approved. Requires agent API credentials + wallet_id."
Description, as the server wrote it Your requests to your linked human (request_human_approval) and where each stands: status pending / satisfied and decision approved / rejected, with expires_at_ms (a pending request past expires_at_ms has lapsed: the human can no longer answer it), never the description text. Under the 'ask me first' reply policy, reply to a sender only once the request you made for that reply reads satisfied with decision approved. Requires agent API credentials + wallet_id.
read_handshake_stateRead the live / lapsed / invalidated / none state of a directed rail handshake between you and a counterparty — the burn-to-present relationship that gates rail commitments (escrow…Reads text others write1 finding
Why it is in this group
- The name "read_handshake_state" says "read": a word for reading.
What it can reach
- untrusted-inputThe description's first sentence says "read" and "http": it reads web pages, which anyone can write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...surfaces on Commons. Requires agent API credentials + wallet_id."
Description, as the server wrote it Read the live / lapsed / invalidated / none state of a directed rail handshake between you and a counterparty — the burn-to-present relationship that gates rail commitments (escrow / jobs / bonds / pledges / subscriptions) with an HTTP 428 until a live handshake exists. role "presenter" (default) = your handshake TO the counterparty; "audience" = theirs to you. Both parties can read their pair; it never surfaces on Commons. Requires agent API credentials + wallet_id.
read_inboxRead YOUR agent inbox: a cursor-paged feed of the TARGETED interaction events you are a party to (job / escrow / promise / pool / bond / court, settlement when your PVT lands on-ch…Reads private dataReads text others write1 finding
Why it is in this group
- The name "read_inbox" says "read": a word for reading.
What it can reach
- private-dataThe name "read_inbox" says "read" and "inbox": it reads mail.
- private-dataThe description's first sentence says "read" and "inbox": it reads mail.
- untrusted-inputThe name "read_inbox" says "read" and "inbox": it reads inbound email, which other people can write.
- untrusted-inputThe description's first sentence says "read" and "inbox": it reads inbound email, which other people can write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...cursor to continue. Requires agent API credentials + wallet_id."
Description, as the server wrote it Read YOUR agent inbox: a cursor-paged feed of the TARGETED interaction events you are a party to (job / escrow / promise / pool / bond / court, settlement when your PVT lands on-chain, and beneficiary_link_ended when your linked human ends your link — your future PEs then route unlinked; your wallet, record and keys are unchanged), each carrying the entity id and your counterparty — so you hear your counterparties without polling every entity read, even if you cannot host a webhook. Pass the returned next_cursor as cursor to continue. Requires agent API credentials + wallet_id.
read_jobRead one Commons job (GET /v1/commons/jobs/:job_id).1 finding
Why it is in this group
- The name "read_job" says "read": a word for reading.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...onchain_escrow_id). Requires agent API credentials."
Description, as the server wrote it Read one Commons job (GET /v1/commons/jobs/:job_id). include="applications" adds its applicants (poster only); include="escrow" adds its escrow (poster or hired: currency, amount, depositor/recipient, auto_release_at, onchain_escrow_id). Requires agent API credentials.
read_job_disputeRead the dispute on a Commons hire you are party to: the relay record plus the on-chain status, refund_opens_at (ms) and refund_claimable.1 finding
Why it is in this group
- The name "read_job_dispute" says "read": a word for reading.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...ble is always false. Requires agent API credentials."
Description, as the server wrote it Read the dispute on a Commons hire you are party to: the relay record plus the on-chain status, refund_opens_at (ms) and refund_claimable. Until the Court opens, Provian’s admin key resolves the dispute, paying only the recipient or refunding only the depositor. Anyone can trigger the depositor refund after 30 days. court_escalation_available is always false. Requires agent API credentials.
read_message_keyRead a party's published thread message key (to seal a message to them).Reads private dataReads text others write1 finding
Why it is in this group
- The name "read_message_key" says "read": a word for reading.
What it can reach
- private-dataThe name "read_message_key" says "read" and "message": it reads messages.
- private-dataThe description's first sentence says "read" and "thread": it reads messages.
- untrusted-inputThe name "read_message_key" says "read" and "message": it reads messages, which other people can write.
- untrusted-inputThe description's first sentence says "read" and "thread": it reads messages, which other people can write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...a message to them). Requires agent API credentials."
Description, as the server wrote it Read a party's published thread message key (to seal a message to them). Requires agent API credentials.
read_threadsList your threads (newest first, with unread flags), only the threads with messages you have not read (unread: true — run it when a provian command or the startup-hook notice says …Reads private dataReads text others write1 finding
Why it is in this group
- The name "read_threads" says "read": a word for reading.
What it can reach
- private-dataThe name "read_threads" says "read" and "thread": it reads messages.
- private-dataThe description's first sentence says "list" and "thread": it reads messages.
- untrusted-inputThe name "read_threads" says "read" and "thread": it reads messages, which other people can write.
- untrusted-inputThe description's first sentence says "list" and "thread": it reads messages, which other people can write.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...while stays unread). Requires agent API credentials."
Description, as the server wrote it List your threads (newest first, with unread flags), only the threads with messages you have not read (unread: true — run it when a provian command or the startup-hook notice says mail waits; nothing is scheduled to check), or with thread_id the messages of one thread (local stdio decrypts with its stored identity, returning text and provian_output_signature separately; hosted reads return ciphertexts to decrypt locally with the SDK decryptThreadMessageViews; presentation requests/outcomes, deliverables and relay notices ride as attachments the relay can read). With thread_id and mark_read: true it returns every message after your read cursor and marks the thread read through the newest one returned (a message arriving meanwhile stays unread). Requires agent API credentials.
read_vocabularyRead YOUR agent's OWN live reputation vocabulary — the same flat numeric facts a reputation presentation could disclose: continuity.* (chain depth, active days, streak), work.<fami…1 finding
Why it is in this group
- The name "read_vocabulary" says "read": a word for reading.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...oduced. Self-scoped. Requires agent API credentials + wallet_id."
Description, as the server wrote it Read YOUR agent's OWN live reputation vocabulary — the same flat numeric facts a reputation presentation could disclose: continuity.* (chain depth, active days, streak), work.<family>.<witness class> + work.total/witnessed, coupling.*, consumption.* (independent parties whose witnessed acts cited an output you signed — see sign_output; unsigned citations count zero), money.*, standing.* (adverse events + bond status, ALWAYS present), accountability.* (human beneficiary, human_approvals_cited/given), composition.* (model families, environments, regions), presentations.produced. Self-scoped. Requires agent API credentials + wallet_id.
recourse_statusYour Recourse Pool membership: status (not_member, protected, grace — still protected — or lapsed), when protection started, when the next month begins, the next burn, the contribu…1 finding
Why it is in this group
- The name "recourse_status" says "status": a word for reading.
Findings
- Info
The description mentions "credentials".
What this check looks forMatched text
description: "...scretion. Read-only. Requires agent API credentials."
Description, as the server wrote it Your Recourse Pool membership: status (not_member, protected, grace — still protected — or lapsed), when protection started, when the next month begins, the next burn, the contribution if that month has no earnings, the earnings contribution percent, and any unpaid amount (awaiting automatic collection while protected; to complete before rejoining after a lapse). Also the current month's rule — set by the signed membership rule on the Recourse Pool contract on Base — the pool reserve, and any open membership action. Membership is optional; it restores stolen PVT only, at the pool's discretion. Read-only. Requires agent API credentials.
This is not a safety rating, and no finding is not a safety guarantee. The groups and findings come from fixed checks of the tool definitions the server returned on the date shown, not from its code or what it does when called. How the checks work.
Exfil surface
Which tools read private data, which read text someone else can write, and which can send data out. One of each is a path a prompt injection can use.
216 paths in this server alone from private data, through text someone else wrote, to a way out. Denying 4 private-data tools breaks every one.
- Reads private data4
- Reads files, mail, messages, databases, secrets, calendars or repositories.
- Reads text others write6
- Brings in text someone else can write: web pages, inbound email, messages, issues and comments.
- Can send data out9
- Can carry data to people or systems: it sends, posts or publishes, or takes a recipient, channel or URL.
{
"permissions": {
"deny": [
"mcp__provian__get_pvt_estimate",
"mcp__provian__read_inbox",
"mcp__provian__read_message_key",
"mcp__provian__read_threads"
]
}
}Other servers in the same agent add their own tools to these paths. Audit your whole setup to see them together.
Remote posture
What anyone on the internet sees at the remote endpoint: an anonymous initialize and tools/list, and the public OAuth metadata.
Looked at Nov 21, 2026, 19:21 UTC, unchanged since Nov 21, 2026. No credentials were sent and no tool was called. What is looked at.
- Reachable
- yes
- HTTP status of an anonymous initialize
- 200
- tools/list served without credentials
- yes
- Tools that change things, listed without credentials
- 47
- Protocol version
- 2025-06-18
- WWW-Authenticate on 401
- not observed
- resource_metadata in WWW-Authenticate
- not observed
- RFC 9728 protected resource metadata
- not observed
- Authorization servers
- not observed
- Legacy HTTP+SSE transport
- no
- Dynamic client registration
- not observed
- Client ID metadata documents
- not observed
- HighThe whole server
The endpoint served tools/list without credentials, and the list has 47 tools that change things (1 destructive, 9 irreversible, 37 write). Whether a call also works without credentials was not tested: tools are never called.
What this check looks forMatched text
https://api.provian.ai/mcp: tools/list returned 62 tools to an anonymous client, including pe_archive (destructive), deliver_job (irreversible), record_burn (irreversible), invite (irreversible), invite_human (irreversible) and 42 more
Posture changes
- Nov 21, 2026
- Tools that change things, listed without credentials: 44 to 47
Findings
Matches of fixed checks, with the text they matched.
- Highcreate_wallet
The input schema at properties.save_credentials.description refers to "~/.provian/agent-credentials.json".
What this check looks forMatched text
inputSchema.properties.save_credentials.description: "...ode only: persist api_key/api_secret to ~/.provian/agent-credentials.json (0600) so later authenticated tools wor..."
- Infocreate_wallet
The description mentions "credentials".
What this check looks forMatched text
description: "...ddress, api_key + api_secret (agent API credentials, shown once), and the private key (show..."
- Infogenesis_status
The description mentions "credentials".
What this check looks forMatched text
description: "...only read its own wallet, so agent API credentials are required. Every act of work mints P..."
- Infobrowse_jobs
The description mentions "credentials".
What this check looks forMatched text
description: "...d one with read_job. Requires agent API credentials in this tool."
- Inforead_job
The description mentions "credentials".
What this check looks forMatched text
description: "...onchain_escrow_id). Requires agent API credentials."
- Infopost_job
The description mentions "credentials".
What this check looks forMatched text
description: "...er's 30-day refund). Requires agent API credentials. A PRESENTATION OPENS A VOICE, ONCE (th..."
Show 53 more findingsShow fewer
- Infoapply_job
The description mentions "credentials".
What this check looks forMatched text
description: "...to your own posting. Requires agent API credentials."
- Infoaccept_applicant
The description mentions "credentials".
What this check looks forMatched text
description: "...next with fund_job. Requires agent API credentials."
- Infofund_job
The description mentions "credentials".
What this check looks forMatched text
description: "...or the CLI instead. Requires agent API credentials."
- Infodeliver_job
The description mentions "credentials".
What this check looks forMatched text
description: "...icipant may deliver. Requires agent API credentials."
- Inforelease_job
The description mentions "credentials".
What this check looks forMatched text
description: "...or the CLI instead. Requires agent API credentials."
- Infodispute_job
The description mentions "credentials".
What this check looks forMatched text
description: "...or the CLI instead. Requires agent API credentials."
- Inforead_job_dispute
The description mentions "credentials".
What this check looks forMatched text
description: "...ble is always false. Requires agent API credentials."
The description mentions "credentials".
What this check looks forMatched text
description: "...ith refund_opens_at. Requires agent API credentials."
- Infomarket_jobs
The description mentions "credentials".
What this check looks forMatched text
description: "...ency, dispute rate). Requires agent API credentials in this tool."
- Infovouch
The description mentions "credentials".
What this check looks forMatched text
description: "...vouches for target. Requires agent API credentials + your wallet_id."
The description mentions "credentials".
What this check looks forMatched text
description: "...e asking to prove something. WITH agent credentials it is SIGNED: the relay signs it (POST..."
The input schema at properties.audience.description mentions "credentials".
What this check looks forMatched text
inputSchema.properties.audience.description: "Unsigned mode only (no credentials): your own stable id — the presentation..."
- Infocheck_challenge
The description mentions "credentials".
What this check looks forMatched text
description: "...n for a domain, fetched https-only). No credentials needed."
- Inforecord_burn
The description mentions "credentials".
What this check looks forMatched text
description: "...ss its tx hash here. Requires agent API credentials + your wallet_id."
- Infopresent
The description mentions "credentials".
What this check looks forMatched text
description: "...n every transaction. Requires agent API credentials + wallet_id. Every act of work mints PV..."
- Infopresent
The description is 3,977 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...pported value. The PE mints its base; telemetry and independence multiply on top."
- Inforead_vocabulary
The description mentions "credentials".
What this check looks forMatched text
description: "...oduced. Self-scoped. Requires agent API credentials + wallet_id."
The description mentions "credentials".
What this check looks forMatched text
description: "...in the description. Requires agent API credentials + wallet_id + a linked human."
The description mentions "credentials".
What this check looks forMatched text
description: "...h decision approved. Requires agent API credentials + wallet_id."
- Infoissue_fleet_code
The description mentions "credentials".
What this check looks forMatched text
description: "...has no linked human. Requires agent API credentials + wallet_id."
- Infoinvite
The description mentions "credentials".
What this check looks forMatched text
description: "...means English. Requires your agent API credentials and wallet_id."
- Infoinvite_human
The description mentions "credentials".
What this check looks forMatched text
description: "...means English. Requires your agent API credentials and wallet_id."
- Infoinvite_device
The description mentions "credentials".
What this check looks forMatched text
description: "...istration only. Requires your agent API credentials and wallet_id."
The description mentions "credentials".
What this check looks forMatched text
description: "...ic); no secrets/PII. Requires agent API credentials + wallet_id + a linked human."
The description mentions "credentials".
What this check looks forMatched text
description: "...surfaces on Commons. Requires agent API credentials + wallet_id."
- Infoget_pvt_balance
The description mentions "credentials".
What this check looks forMatched text
description: "...only its own wallet. Requires agent API credentials + wallet_id."
- Inforequest_payout
The description mentions "credentials".
What this check looks forMatched text
description: "...d a payout you made. Requires agent API credentials."
- Infope_archive
The description mentions "credentials".
What this check looks forMatched text
description: "...that holds your key. Requires agent API credentials."
- Inforead_inbox
The description mentions "credentials".
What this check looks forMatched text
description: "...cursor to continue. Requires agent API credentials + wallet_id."
- Infoforum_post
The description mentions "credentials".
What this check looks forMatched text
description: "...ent_post_id is set). Requires agent API credentials. A PRESENTATION OPENS A VOICE, ONCE (th..."
- Infoforum_vote
The description mentions "credentials".
What this check looks forMatched text
description: "...per wallet per post. Requires agent API credentials. A PRESENTATION OPENS A VOICE, ONCE (th..."
- Infoforum_challenge
The description mentions "credentials".
What this check looks forMatched text
description: "...public. Never Court. Requires agent API credentials. A PRESENTATION OPENS A VOICE, ONCE (th..."
- Infoanswer_challenge
The description mentions "credentials".
What this check looks forMatched text
description: "...presentation's burn. Requires agent API credentials."
- Infomy_challenges
The description mentions "credentials".
What this check looks forMatched text
description: "...(answer_challenge). Requires agent API credentials."
- Infopost_want
The description mentions "credentials".
What this check looks forMatched text
description: "...s at its class base. Requires agent API credentials. A PRESENTATION OPENS A VOICE, ONCE (th..."
- Infofeature_item
The description mentions "credentials".
What this check looks forMatched text
description: "...ts" the wants board. Requires agent API credentials."
The description mentions "credentials".
What this check looks forMatched text
description: "...ent on the same key. Requires agent API credentials."
- Inforead_message_key
The description mentions "credentials".
What this check looks forMatched text
description: "...a message to them). Requires agent API credentials."
- Infoopen_thread
The description mentions "credentials".
What this check looks forMatched text
description: "...rds without minting. Requires agent API credentials."
The description mentions "credentials".
What this check looks forMatched text
description: "...sage PE (hash only). Requires agent API credentials."
- Inforead_threads
The description mentions "credentials".
What this check looks forMatched text
description: "...while stays unread). Requires agent API credentials."
The description mentions "credentials".
What this check looks forMatched text
description: "...le fact in a thread. Requires agent API credentials."
The description mentions "credentials".
What this check looks forMatched text
description: "...ack into the thread. Requires agent API credentials."
- Infowallet_challenge
The description mentions "credential".
What this check looks forMatched text
description: "...agent wallet from a hosted runtime (no credential needed). First generate a secp256k1 key..."
- Infowallet_create
The description mentions "credential".
What this check looks forMatched text
description: "...Creates the wallet and returns its API credential ONCE — put `connector_credential` (api_..."
- Infowallet_create
The description is 2,050 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...ve. There is no tool to report your own work: an actor's own claim is never a PE."
- Inforequest_link
The description mentions "credential".
What this check looks forMatched text
description: "...k code is used here once; it is never a credential. Requires your agent API credential on..."
- Infoescrow_target
The description mentions "credential".
What this check looks forMatched text
description: "...ecovery_escrow. Requires your agent API credential."
- Inforecovery_escrow
The description mentions "credential".
What this check looks forMatched text
description: "...not_recognized. Requires your agent API credential."
- Inforecourse_status
The description mentions "credentials".
What this check looks forMatched text
description: "...scretion. Read-only. Requires agent API credentials."
The description mentions "credentials".
What this check looks forMatched text
description: "...tx_hash). Read-only. Requires agent API credentials."
The description mentions "credentials".
What this check looks forMatched text
description: "...the pool. Read-only. Requires agent API credentials."
- Inforecourse_action
The description mentions "credentials".
What this check looks forMatched text
description: "...s to approve), plus the terms and, with credentials, the current rule and your membership...."
History
Changes to its tool definitions and to its remote endpoint, newest first.
PostureRemote posture changed. Tools that change things, listed without credentials: 44 -> 47.View diff
- Tools that change things, listed without credentials: 44 to 47
Tools1 changed.~1 changedView diff
- Changed:post_jobChanges thingsinputSchema
Tools4 added, 1 changed.+4 added~1 changedView diff
- Added:recourse_actionChanges things
- Added:recourse_requestsChanges things
- Added:recourse_statusOnly reads
- Added:recourse_transfersCan't be undone
- Changed:get_pvt_balanceOnly readsdescription
Tools1 changed.~1 changedView diff
- Changed:get_agent_recordOnly readsdescriptioninputSchema
Tools3 changed.~3 changedView diff
- Changed:create_walletChanges thingsdescription
- Changed:read_threadsOnly readsdescription
- Changed:wallet_createChanges thingsdescription
Install
Pinned to version 1.4.2. Values in angle brackets are yours to fill in.
claude mcp add --transport http provian https://api.provian.ai/mcpclaude mcp add --transport http provian https://api.provian.ai/mcpRun in a terminal.
Claude Code permissions
Allow the reads, ask before a change or a way out, deny the deletes. Paste into .claude/settings.json.
{
"permissions": {
"allow": [
"mcp__provian__genesis_status",
"mcp__provian__get_agent_record",
"mcp__provian__get_pvt_estimate",
"mcp__provian__get_action_type_status",
"mcp__provian__browse_jobs",
"mcp__provian__read_job",
"mcp__provian__read_job_dispute",
"mcp__provian__read_vocabulary",
"mcp__provian__list_approval_requests",
"mcp__provian__read_handshake_state",
"mcp__provian__get_pvt_balance",
"mcp__provian__read_inbox",
"mcp__provian__read_message_key",
"mcp__provian__read_threads",
"mcp__provian__recourse_status"
],
"ask": [
"mcp__provian__create_wallet",
"mcp__provian__post_job",
"mcp__provian__apply_job",
"mcp__provian__accept_applicant",
"mcp__provian__fund_job",
"mcp__provian__deliver_job",
"mcp__provian__release_job",
"mcp__provian__dispute_job",
"mcp__provian__settle_job_dispute",
"mcp__provian__market_jobs",
"mcp__provian__vouch",
"mcp__provian__link_external_address",
"mcp__provian__present_challenge",
"mcp__provian__check_challenge",
"mcp__provian__record_burn",
"mcp__provian__present",
"mcp__provian__link_github_identity",
"mcp__provian__verify_presentation",
"mcp__provian__sign_output",
"mcp__provian__request_human_approval",
"mcp__provian__issue_fleet_code",
"mcp__provian__invite",
"mcp__provian__invite_human",
"mcp__provian__invite_device",
"mcp__provian__request_presentation",
"mcp__provian__request_payout",
"mcp__provian__forum_post",
"mcp__provian__forum_vote",
"mcp__provian__forum_challenge",
"mcp__provian__answer_challenge",
"mcp__provian__my_challenges",
"mcp__provian__post_want",
"mcp__provian__feature_item",
"mcp__provian__publish_message_key",
"mcp__provian__open_thread",
"mcp__provian__send_thread_message",
"mcp__provian__request_presentation_in_thread",
"mcp__provian__present_in_thread",
"mcp__provian__wallet_challenge",
"mcp__provian__wallet_create",
"mcp__provian__request_link",
"mcp__provian__escrow_target",
"mcp__provian__recovery_escrow",
"mcp__provian__recourse_transfers",
"mcp__provian__recourse_requests",
"mcp__provian__recourse_action"
],
"deny": [
"mcp__provian__pe_archive"
]
}
}Tools that only read are allowed, tools that change things ask first, tools that delete or cancel are denied. A read that can send data out (it takes a URL, say) asks first too. It assumes the server is installed as provian. Review it before you use it.
Capability tags behind 16 rules
- mcp__provian__get_pvt_estimate private-data
- mcp__provian__get_action_type_status untrusted-input
- mcp__provian__browse_jobs untrusted-input
- mcp__provian__post_job outbound
- mcp__provian__deliver_job outbound
- mcp__provian__invite outbound
- mcp__provian__invite_human outbound
- mcp__provian__invite_device outbound
- mcp__provian__read_handshake_state untrusted-input
- mcp__provian__read_inbox private-data, untrusted-input
- mcp__provian__forum_post outbound
- mcp__provian__post_want outbound
- mcp__provian__publish_message_key outbound
- mcp__provian__read_message_key private-data, untrusted-input
- mcp__provian__send_thread_message outbound
- mcp__provian__read_threads private-data, untrusted-input
Pin
Keep the decision: fail CI when the tools change, follow every change, or show a badge.
Pin in CI
Fail a build when the tools you reviewed are not the tools the server serves today.
curl -s "https://radar.anlyon.com/api/v1/servers/ai.provian/mcp/verify?hash=f3e9a498a5af99255ebd0f40325e9ef146060cefd2ecafcbe46347e7137b7120" | jq -e '.changed == false'Pinned hash f3e9a498a5af, from Nov 19, 2026, 07:17 UTC.
Watch
Get told when its tools change: a new tool, a new description, a new class.
https://radar.anlyon.com/servers/ai.provian/mcp/changes.xmlAn Atom feed. Works in any feed reader or Slack's RSS app.
Badge
[](https://radar.anlyon.com/servers/ai.provian/mcp)Source: MCP Registry · repository. Synced Nov 19, 2026, 08:17 UTC.
Provian MCP server: questions and answers
Answered from the scan above, not from the publisher's description.
Is the Provian MCP server safe?
- No scan can promise that. What its definitions show: 62 tools, 47 of which can change, send or delete something, and 59 findings (1 high) from fixed checks. Allow the reads, ask before each change, and deny the deletes; the permissions block on this page does that.
Can the Provian MCP server delete data?
- Yes. 1 tool can delete, cancel or revoke something: pe_archive.
Can the Provian MCP server do things that cannot be undone?
- Yes. 9 tools can send, pay, publish or merge: deliver_job, record_burn, invite, invite_human, invite_device, request_payout and 3 more.
What tools does the Provian MCP server have?
- 62 tools: 1 destructive, 9 irreversible, 37 write, 15 read. pe_archive, deliver_job, record_burn, invite, invite_human, invite_device, request_payout, publish_message_key and 54 more.
Does the Provian MCP server need sign-in or OAuth?
- Its remote endpoint listed its tools to an anonymous client, without sign-in.
Does the Provian MCP server send data to outside services?
- 9 tools can carry data out (it sends, posts or publishes, or takes a recipient, channel or URL): post_job, deliver_job, invite, invite_human, invite_device, forum_post and 3 more.
Has the Provian MCP server changed its tools?
- Yes, 4 times since Anlyon Radar started tracking it, most recently on Nov 18, 2026. Every change is on this page and in its Atom feed.