Skip to content

mailbox.bot MCP server

bot.mailbox/mailboxmailbox.botstreamable-httpv1.0.5Scanned

Hosted mailbox.bot postal MCP. $20/mo mailboxes are live now.

55 tools: 7 can delete or cancel things, 2 do something that can't be undone, 16 change things and 30 only read.

Tools: 7 delete, 2 irreversible, 16 write, 30 read.
  • Reads private data · 24
  • Reads text others write · 37
  • Can send data out · 7
  • 23 findings

Tools listed without sign-in. Remote posture

With Anlyon

7 tools here can delete or cancel things and 2 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.
Govern this server →

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 mailbox.bot MCP server can do

The mailbox.bot MCP server is published by mailbox.bot, a domain the MCP Registry verified. It has 55 tools. It can delete or cancel things (7 tools: send_outbound_mail, cancel_outbound_mail, update_webhook_endpoint, rotate_webhook_endpoint_secret and 3 more), do things that cannot be undone, such as sending, paying or publishing (send_facility_message and create_batch_mail) and create or change data (16 tools: propose_mailbox_md_edit, update_webhook, create_test_outbound_mail, advance_test_outbound_mail and 12 more). 30 tools only read.

24 tools read private data (get_mailbox, list_inbound_forwarding_addresses, list_inbound_mail and 21 more); 37 tools bring in text other people can write (get_mailbox, list_inbound_forwarding_addresses, list_inbound_mail and 34 more); 7 tools can send data out of your system (send_facility_message, update_webhook, send_outbound_mail and 4 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 23 findings (long-description, credentials-mention and tool-shadowing). Each one comes with the text it matched.

Its remote endpoint lists its tools without sign-in, and its tools changed 3 times since Anlyon Radar first saw them.

Last scanned Nov 20, 2026, version 1.0.5.

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.

55 tools

Deletes or cancels 7Claude Code: deny
  • cancel_batch_mailCancel a batch before the facility starts (draft, paid, generating or submitted); credits are refunded in full.Reads text others writeNo findings

    Why it is in this group

    • The name "cancel_batch_mail" says "cancel": a word for removing or ending something.
    • The server marks it destructive (destructiveHint: true).

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    Description, as the server wrote it
    Cancel a batch before the facility starts (draft, paid, generating or submitted); credits are refunded in full. Once the facility is printing (in_progress) it cannot be cancelled.

    Annotations the server declares

    • title: Cancel Batch Mail
    • readOnlyHint: false
    • destructiveHint: true
    • idempotentHint: true
    • openWorldHint: true
  • cancel_outbound_mailCancel a queued outbound mail job before facility printing starts.No findings

    Why it is in this group

    • The name "cancel_outbound_mail" says "cancel": a word for removing or ending something.
    • The server marks it destructive (destructiveHint: true).
    Description, as the server wrote it
    Cancel a queued outbound mail job before facility printing starts. If the mail was funded with prepaid credits, eligible credits are returned to the member ledger. Safe to retry: already-cancelled mail returns cancelled status without creating a duplicate refund. In chat, report cancellation status, returned credits, updated balance, and whether it had already been cancelled. If a transient error occurs, poll the mail status and credits before retrying.

    Annotations the server declares

    • title: Cancel Outbound Mail
    • readOnlyHint: false
    • destructiveHint: true
    • idempotentHint: true
    • openWorldHint: false
  • confirm_batch_mailPay for a draft from the account's prepaid credits and queue it for printing.Reads text others writeNo findings

    Why it is in this group

    • The description's first verb is "pay": a word that usually means money moves, which cannot be taken back.
    • The server marks it destructive (destructiveHint: true).

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    Description, as the server wrote it
    Pay for a draft from the account's prepaid credits and queue it for printing. Debits exactly batch.pricing.total_cents once (repeating returns the current state). Only after the person has seen the proof. The key's policies apply: a key that requires approval cannot pay (the owner confirms in the dashboard), max_daily_pieces counts the whole batch, and max_cost_cents refuses before any debit. A sandbox key pays nothing, and a sandbox batch moves forward only through the sandbox advance call.

    Annotations the server declares

    • title: Confirm Batch Mail
    • readOnlyHint: false
    • destructiveHint: true
    • idempotentHint: true
    • openWorldHint: true
  • rotate_webhook_endpoint_secretRotate the signing secret; requires confirm_rotation:true.Reads text others writeNo findings

    Why it is in this group

    • Nothing in its name or description says what it does, so it is counted as a write.
    • The server marks it destructive (destructiveHint: true).

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    Description, as the server wrote it
    Rotate the signing secret; requires confirm_rotation:true. Old signatures overlap for 24 hours while the receiver's secure secret configuration is updated. Requires webhook.manage. Uses the same bounded control plane and tenant/agent/Live-Sandbox authorization as /api/v1/webhooks/endpoints. A repeated idempotency_key with the same input replays the stored result.

    Annotations the server declares

    • title: Rotate Webhook Endpoint Secret
    • readOnlyHint: false
    • destructiveHint: true
    • idempotentHint: true
    • openWorldHint: true
  • send_outbound_mailSubmit a document for printing and postal mailing by the facility.Reads text others writeCan send data out1 finding

    Why it is in this group

    • The name "send_outbound_mail" says "send": a word for sending, and what is sent cannot be taken back.
    • The server marks it destructive (destructiveHint: true).

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    • outboundThe name "send_outbound_mail" says "send": it sends or publishes something to other people.

    Findings

    • Info

      The description is 2,533 characters long, over the 2,000 this check looks for.

      Matched text

      description ends: "...l_thread_id so lineage stays explicit. Explicit Business mail runs are REST-only."

      What this check looks for
    Description, as the server wrote it
    Submit a document for printing and postal mailing by the facility. Supported formats: PDF, DOCX, JPG, PNG, TXT, CSV. The document is stored securely and printed by the facility operator. USPS First-Class Mail has a published $3.00 one-page baseline. Extra pages add configured printing plus any postage increase from weight. If an account's normal first-page formula is lower, cost_breakdown.first_class_baseline_adjustment_cents itemizes the difference separately from carrier postage. Published default pricing for other classes is $2.50 handling + $0.40/page B&W printing + carrier postage/rate; FedEx Express Saver, 2Day and Overnight are $5.00 below their otherwise-calculated carrier baselines: fedex_express itemizes service_adjustment_cents: -500, FedEx 2Day applies a fixed $13.00 customer price reduction (service_adjustment_cents: -1300), and FedEx Overnight adds a fixed $13.00 adjustment (service_adjustment_cents: 1300); fedex_ground carries a fixed $5.00 markup (service_adjustment_cents: 500). Color pages cost $0.70/page total by default before handling and postage: $0.40/page base B&W printing plus a $0.30/page additive color upgrade. In cost_breakdown, color_printing_per_page_cents is the total color-page printing price (70 cents by default), while color_surcharge_per_page_cents is the additive upgrade only (30 cents by default). Account-specific pricing can override these defaults; cost_breakdown is authoritative. IMPORTANT: With a production key (sk_agent_), this spends the human member's prepaid mailbox.bot credits. Agents never access Stripe, card data, or Auto-Fill settings. If the signed-in human separately enabled Agent Auto-Fill, an eligible live order may trigger a bounded server-managed reload. Use dry_run=true to preview required credits before committing, or requires_approval=true to defer the credit debit until human approval. Sandbox keys (sk_agent_test_) skip credit debits and facility fulfillment. Responses include human_review with send-to address, return address, mail class, document details, preview URL when available, cost, safeguards, and next step; show that to the human before live funded sends. tracking_number is required for priority, certified, certified_return_receipt, FedEx, and UPS mail classes. USPS first_class does not include carrier tracking by default; tracking_number may be null. Optionally attach the outbound mail to inbound context with inbound_capture_id and postal_mail_thread_id so lineage stays explicit. Explicit Business mail runs are REST-only.

    Annotations the server declares

    • title: Send Outbound Mail
    • readOnlyHint: false
    • destructiveHint: true
    • idempotentHint: false
    • openWorldHint: true
  • send_text_letterMail a letter written as plain text, for letters composed in chat.Reads text others writeCan send data outNo findings

    Why it is in this group

    • The name "send_text_letter" says "send": a word for sending, and what is sent cannot be taken back.
    • The server marks it destructive (destructiveHint: true).

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    • outboundThe name "send_text_letter" says "send": it sends or publishes something to other people.
    Description, as the server wrote it
    Mail a letter written as plain text, for letters composed in chat. mailbox.bot sets the text exactly as given on US Letter pages (Times 12 pt, 1-inch margins; no letterhead, logo or signature image) and submits it through the same outbound mail service, so the same scopes, approval, cost caps and proof apply and staff print and mail it. dry_run: true returns the exact cost without sending; with a live key this spends the account owner's prepaid credits, and a key that requires approval holds the letter until the owner approves it in the dashboard. Requires the current mailbox_md_version. Up to 15,000 characters of Latin text; plain text only, with no PDF, letterhead or images.

    Annotations the server declares

    • title: Send Text Letter
    • readOnlyHint: false
    • destructiveHint: true
    • idempotentHint: false
    • openWorldHint: true
  • update_webhook_endpointUpdate name, URL, events, keyword rules or active/paused status with expected_revision.Reads text others write1 finding

    Why it is in this group

    • The name "update_webhook_endpoint" says "update": a word for a change.
    • The server marks it destructive (destructiveHint: true).

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.

    Findings

    • Info

      The input schema at properties.url.description mentions "credentials".

      Matched text

      inputSchema.properties.url.description: "Public HTTPS URL on port 443 (no credentials, fragment or private host) that receive..."

      What this check looks for
    Description, as the server wrote it
    Update name, URL, events, keyword rules or active/paused status with expected_revision. Agent/environment scope cannot be changed. Requires webhook.manage. Uses the same bounded control plane and tenant/agent/Live-Sandbox authorization as /api/v1/webhooks/endpoints. A repeated idempotency_key with the same input replays the stored result.

    Annotations the server declares

    • title: Update Webhook Endpoint
    • readOnlyHint: false
    • destructiveHint: true
    • idempotentHint: true
    • openWorldHint: true
Can't be undone 2Claude Code: ask
  • create_batch_mailCreate a batch draft from a template link and a recipient list; nothing is charged.Reads text others writeCan send data outNo findings

    Why it is in this group

    • The name "create_batch_mail" says "create": a word for a change.
    • The input addresses people (input.recipients): what it sends cannot be taken back.

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    • outboundThe name "create_batch_mail" says "create" and "mail": it writes something other people read.
    • outboundThe input takes a destination (input.recipients): what the tool writes goes there.
    • outboundThe input takes a URL (input.qr.url): a request to a URL can carry data to whoever runs it.
    Description, as the server wrote it
    Create a batch draft from a template link and a recipient list; nothing is charged. template_url is a public https:// link to a PDF at the mail type's size, or to a PNG/JPG postcard graphic that is set on a page at that size (design 4×6 in at 300 dpi = 1200×1800 px). Recipients go inline (an array of {name, line1, line2, city, state, zip}; company accepted) or as recipients_url, a CSV link. Malformed rows and duplicates are skipped and listed, never charged. A one-page postcard gets an address side printed with the return address. Field kits, all optional: every other recipient key (or a nested fields object) is a field, printed by {{tag}} in message or qr.url (fallback: {{tag|text}}) and into PDF form fields of the same name; qr puts a code unique to each piece on it (link opens qr.url, confirm opens a one-button mailbox.bot page, verify proves the recipient gets mail at the address: a QR code plus a short code the person enters where they are signed in, your app via POST /v1/batch-mail/verify with their ref, or read back to the owner for the dashboard; set qr.url to that page and a scan opens it with the code filled in), with scans and confirmations per piece on the batch record and as batch.piece.scanned / batch.piece.confirmed webhooks (a "ref" key on each recipient comes back with them); stock weatherproof, one printed insert (insert_url + insert_stock) and mail_on (a date to mail on) as well. The result carries the exact price, the first recipients and any image warnings. Nothing is charged until the batch is confirmed, and a one-hour proof link shows the template and a sample address label first. Requires the current mailbox_md_version. A key that requires approval saves the draft for the account owner to pay in the dashboard. Postcards from 1 recipient (a $2.00 small-batch fee under 10); letters, flyers and pamphlets from 10; up to 10,000.

    Annotations the server declares

    • title: Create Batch Mail
    • readOnlyHint: false
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: true
  • send_facility_messageSend a message to the operator at your mailbox facility.Reads text others writeCan send data outNo findings

    Why it is in this group

    • The name "send_facility_message" says "send": a word for sending, and what is sent cannot be taken back.

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    • outboundThe name "send_facility_message" says "send": it sends or publishes something to other people.
    Description, as the server wrote it
    Send a message to the operator at your mailbox facility. Facility routing is automatic. Messages appear in the shared conversation visible to you, the renter, and the facility. Optionally link the message to a specific package or action request for context.

    Annotations the server declares

    • title: Send Facility Message
    • readOnlyHint: false
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: true
Changes things 16Claude Code: ask
  • advance_test_batch_mailSandbox keys only: simulate the facility one step (submitted → in_progress → completed).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
    Sandbox keys only: simulate the facility one step (submitted → in_progress → completed). Completion marks every piece mailed and fires batch.completed to the agent's webhook, exactly as the facility would. With qr_event and piece it records a scan or a confirmation on that piece of a sandbox batch with QR codes instead, as if a phone scanned it, and the batch.piece.* webhooks fire. Live batches are advanced by the facility, never by the API.

    Annotations the server declares

    • title: Advance Test Batch Mail
    • readOnlyHint: false
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: false
  • advance_test_outbound_mailAdvance a test_mode outbound mail record one lifecycle step and queue the matching webhook.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
    Advance a test_mode outbound mail record one lifecycle step and queue the matching webhook. submitted becomes ready with simulated pages/envelope photos; ready becomes mailed with carrier, dispatch method, receipt photo, and tracking when the selected service includes tracking; mailed becomes delivered.

    Annotations the server declares

    • title: Advance Test Outbound Mail
    • readOnlyHint: false
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: false
  • create_test_outbound_mailCreate a sandbox outbound mail record without uploading a real document.Can send data outNo findings

    Why it is in this group

    • The name "create_test_outbound_mail" says "create": a word for a change.

    What it can reach

    • outboundThe name "create_test_outbound_mail" says "create" and "mail": it writes something other people read.
    Description, as the server wrote it
    Create a sandbox outbound mail record without uploading a real document. The record is always test_mode=true, cost_cents=0, includes estimated_live_cost_cents and cost_breakdown, and queues a mail.submitted webhook. USPS First-Class Mail has a published $3.00 one-page baseline. Extra pages add configured printing plus any postage increase from weight. If an account's normal first-page formula is lower, cost_breakdown.first_class_baseline_adjustment_cents itemizes the difference separately from carrier postage. Color pages cost $0.70/page total by default before handling and postage: $0.40/page base B&W printing plus a $0.30/page additive color upgrade. In cost_breakdown, color_printing_per_page_cents is the total color-page printing price (70 cents by default), while color_surcharge_per_page_cents is the additive upgrade only (30 cents by default). Account-specific pricing can override these defaults; cost_breakdown is authoritative. FedEx and UPS estimates use the same configured origin and destination zone/region logic as production; fedex_express, fedex_2day and fedex_overnight are $5.00 below their baselines (fedex_express reports service_adjustment_cents: -500), FedEx 2Day applies a fixed $13.00 customer price reduction and reports service_adjustment_cents: -1300, FedEx Overnight adds a fixed $13.00 adjustment and reports service_adjustment_cents: 1300, and fedex_ground carries a fixed $5.00 markup and reports service_adjustment_cents: 500. Use with a sandbox key to rehearse outbound workflows before sending real physical mail.

    Annotations the server declares

    • title: Create Test Outbound Mail
    • readOnlyHint: false
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: false
  • create_webhook_endpointCreate a webhook: name, public HTTPS URL, events (inbound.received, inbound.action.requested, inbound.action.completed, inbound.pages_ready, inbound.keywords_matched) and optional …Reads text others writeCan send data out1 finding

    Why it is in this group

    • The name "create_webhook_endpoint" says "create": a word for a change.

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    • outboundThe input takes a URL (input.url): a request to a URL can carry data to whoever runs it.

    Findings

    • Info

      The input schema at properties.url.description mentions "credentials".

      Matched text

      inputSchema.properties.url.description: "Public HTTPS URL on port 443 (no credentials, fragment or private host) that receive..."

      What this check looks for
    Description, as the server wrote it
    Create a webhook: name, public HTTPS URL, events (inbound.received, inbound.action.requested, inbound.action.completed, inbound.pages_ready, inbound.keywords_matched) and optional keyword rules (literal terms, any/all, exclusions). Active on save. Returns the whsec_ signing secret once, for the receiver's secure configuration. Agent keys can only create endpoints for their own agent_id and environment; member keys may use agent_id:null for account scope. Requires webhook.manage. Uses the same bounded control plane and tenant/agent/Live-Sandbox authorization as /api/v1/webhooks/endpoints. A repeated idempotency_key with the same input replays the stored result.

    Annotations the server declares

    • title: Create Webhook Endpoint
    • readOnlyHint: false
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • estimate_batch_mailPrice a batch before any file exists.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.
    • The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
    Description, as the server wrote it
    Price a batch before any file exists. Per piece: printing, card stock or paper, the envelope where one is used, the address label, handling and USPS postage. First-Class: 4×6 postcard $1.05, 6×9 postcard $1.35 (both sides in full color; a one-page design gets an address side printed with the return address), letter $1.15 (1 sheet) or $1.35 (2–3 sheets), flyer $1.15, pamphlet $1.95 (4–8 pages) or $2.75 (9–16 pages). Marketing Mail (200+ recipients, not open yet): postcards 85¢, letters and flyers 89¢, pamphlets $1.75. Color on letters, flyers and pamphlets adds 16¢ per color page; duplex is included (two pages per sheet). Volume: 5 % off at 500+, 10 % at 1,000+. No piece is priced below its minimum, our cost to make and mail it plus 5¢: two-sided color, extra sheets, the volume discount and account prices all stop there, and the estimate says when it applied. Postcards from 1 piece (a $2.00 small-batch fee under 10); letters, flyers and pamphlets from 10 (fewer: POST /v1/mail); up to 10,000 per batch. The estimate is the exact amount confirm debits. Sandbox keys get the same numbers with credits_required_cents 0.

    Annotations the server declares

    • title: Estimate Batch Mail
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • propose_agent_inbox_handlingDeprecated: superseded by the tool named in _meta.successor.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 "credential".

      Matched text

      description: "...s, current member approval, assignment, credential environment and SQL authorization are s..."

      What this check looks for
    • Warn

      The description names a tool "in" that this server does not list.

      Matched text

      description: "Deprecated: superseded by the tool named in _meta.successor. Still served unchanged..."

      What this check looks for
    Description, as the server wrote it
    Deprecated: superseded by the tool named in _meta.successor. Still served unchanged during the alias window. Propose through /v1/agent-inbox/:id/handling with dedicated agent.inbox.propose and read scopes. Requests carry confirmed:false and destructive_confirmed:false and the item's current handling version; humans approve before the paired facility queue or alert. Forward requires complete US destination and mail class; discard needs subsequent owner/staff confirmation. Sample open_and_scan is TEST training, not paid OCR. A retry with the identical idempotency_key, body and original expected_version replays safely. Cannot self-approve, dispatch postage, charge, destroy or complete work. Required scopes: agent.inbox.read, agent.read, agent.inbox.propose. Agent-scoped bearer key required; member/facility keys cannot impersonate an agent. Rollout switches, current member approval, assignment, credential environment and SQL authorization are shared with REST.

    Annotations the server declares

    • title: Propose Agent Inbox Handling
    • readOnlyHint: false
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • propose_mailbox_md_editPropose a change to the renter's MAILBOX.md, with a reason.No findings

    Why it is in this group

    • The name "propose_mailbox_md_edit" says "edit": a word for a change.
    Description, as the server wrote it
    Propose a change to the renter's MAILBOX.md, with a reason. The renter sees the suggestion in their dashboard and can accept, reject, or modify it.

    Annotations the server declares

    • title: Propose MAILBOX.md Edit
    • readOnlyHint: false
    • destructiveHint: false
    • idempotentHint: false
    • openWorldHint: false
  • quote_inbound_forwardPrice every forwarding class for an item to a US ZIP through GET /v1/inbound-items/:id/forward-quote: cost_cents = carrier postage from the facility plus handling, with the breakdo…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.
    • The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
    Description, as the server wrote it
    Price every forwarding class for an item to a US ZIP through GET /v1/inbound-items/:id/forward-quote: cost_cents = carrier postage from the facility plus handling, with the breakdown, tracking notes and bill timing. A proposal for the chosen class echoes it as action.expected_quote { cost_cents, billable, max_cents: cost_cents }; on a sample or test item cost_cents is 0 and live_cost_cents shows the real price. Key scope: inbound.item.read (agent.inbox.read also works). Member keys read their own live items; agent keys read items bound or assigned to them, in their key's environment. Proposals are agent-only and always wait for the owner: they never charge credits, buy postage, open, forward or discard anything by themselves. Forward destinations and discard proposals are checked against the agent's structured inbound policy. Image URLs are never issued to agents.

    Annotations the server declares

    • title: Quote Inbound Forward
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • quote_outbound_mailPrice a letter before you have the document.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.
    • The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
    Description, as the server wrote it
    Price a letter before you have the document. Takes the mail class, page count, color and, for FedEx, the recipient ZIP; returns mailbox.bot's published price for a new account with its breakdown, in US dollars. Needs no account and creates nothing. The dry run of the real document is the exact price for an account.

    Annotations the server declares

    • title: Quote Outbound Mail
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • replay_webhook_deliveryExplicitly retry a failed delivery after correcting the receiver.Reads text others writeNo findings

    Why it is in this group

    • Nothing in its name or description says what it does, so it is counted as a write.

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    Description, as the server wrote it
    Explicitly retry a failed delivery after correcting the receiver. Preserves event_id for deduplication and rechecks the current endpoint revision. Does not replay successful deliveries or approve facility handling. Requires webhook.manage. Uses the same bounded control plane and tenant/agent/Live-Sandbox authorization as /api/v1/webhooks/endpoints. A repeated idempotency_key with the same input replays the stored result.

    Annotations the server declares

    • title: Replay Webhook Delivery
    • readOnlyHint: false
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • report_agent_inbox_outcomeDeprecated: no successor; retired when the alias window closes.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".

      Matched text

      description: "...s, current member approval, assignment, credential environment and SQL authorization are s..."

      What this check looks for
    Description, as the server wrote it
    Deprecated: no successor; retired when the alias window closes. Still served unchanged until then. Report actual external-worker processing through /v1/agent-inbox/:id/acknowledgments. Takes the exact fetched context_version, outcome and non-sensitive run_id in request. Needs agent.inbox.report plus both read scopes. A retry with the same idempotency_key and request replays the stored result. Reports cannot approve or complete facility handling. Required scopes: agent.inbox.read, agent.read, agent.inbox.report. Agent-scoped bearer key required; member/facility keys cannot impersonate an agent. Rollout switches, current member approval, assignment, credential environment and SQL authorization are shared with REST.

    Annotations the server declares

    • title: Report Agent Inbox Outcome
    • readOnlyHint: false
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • request_inbound_actionPropose a scan, forward or discard through POST /v1/inbound-items/:id/actions.No findings

    Why it is in this group

    • The name "request_inbound_action" says "request": a word for a change.
    Description, as the server wrote it
    Propose a scan, forward or discard through POST /v1/inbound-items/:id/actions. Agent keys only; the proposal always waits for the owner's approval (awaiting_member_approval: true) and never charges credits or moves mail by itself. Takes the item's current version as action.expected_version, its current quote as action.expected_quote (the item's quotes.scan, or the forward quote for the chosen class; required for scan and forward, refused with the fresh quote otherwise), an idempotency_key that makes retries safe, and for forward a complete US destination and mail_class (first_class requires untracked_acknowledged: true; packages cannot be forwarded yet). Forward destinations and discard proposals are checked against the agent's inbound_policy server-side. One open action per item. Key scope: inbound.item.action (agent.inbox.propose also works). Member keys read their own live items; agent keys read items bound or assigned to them, in their key's environment. Proposals are agent-only and always wait for the owner: they never charge credits, buy postage, open, forward or discard anything by themselves. Forward destinations and discard proposals are checked against the agent's structured inbound policy. Image URLs are never issued to agents.

    Annotations the server declares

    • title: Request Inbound Action
    • readOnlyHint: false
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • seed_agent_inbox_sandboxDeprecated: no successor; retired when the alias window closes.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".

      Matched text

      description: "...s, current member approval, assignment, credential environment and SQL authorization are s..."

      What this check looks for
    Description, as the server wrote it
    Deprecated: no successor; retired when the alias window closes. Still served unchanged until then. Explicitly seed an isolated provider-free two_page_letter or needs_review fixture through /v1/agent-inbox/sandbox/seed. Sandbox agent key only. Exact retry key/scenario reuses the fixture. No member sample, real mail, model call, email, facility task, PMB or charge; these fixtures test reads/reports, not handling. Required scopes: agent.inbox.read. Agent-scoped bearer key required; member/facility keys cannot impersonate an agent. Rollout switches, current member approval, assignment, credential environment and SQL authorization are shared with REST.

    Annotations the server declares

    • title: Seed Agent Inbox Sandbox
    • readOnlyHint: false
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • test_webhook_endpointQueue a synthetic webhook.test delivery (no item) to check the receiver and its signature verification.Reads text others writeNo findings

    Why it is in this group

    • Nothing in its name or description says what it does, so it is counted as a write.

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    Description, as the server wrote it
    Queue a synthetic webhook.test delivery (no item) to check the receiver and its signature verification. Queue admission is not HTTP delivery; the delivery log shows the result. Requires webhook.manage. Uses the same bounded control plane and tenant/agent/Live-Sandbox authorization as /api/v1/webhooks/endpoints. A repeated idempotency_key with the same input replays the stored result.

    Annotations the server declares

    • title: Test Webhook Endpoint
    • readOnlyHint: false
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • test_webhook_endpoint_with_sampleRun the Sample test: opens and scans the member's Mojave sample letter if needed (free, fictional), delivers its real inbound.pages_ready / inbound.keywords_matched events to this …Reads text others writeNo findings

    Why it is in this group

    • The description's first verb is "run": a word for a change.

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    Description, as the server wrote it
    Run the Sample test: opens and scans the member's Mojave sample letter if needed (free, fictional), delivers its real inbound.pages_ready / inbound.keywords_matched events to this webhook, and returns a verdict computed from the receiver's actual HTTP result plus a next step. success proves transport receipt and, when keywords are configured, a keyword match; it never proves agent processing. No mail, charge, email or facility work. Requires webhook.manage. Uses the same bounded control plane and tenant/agent/Live-Sandbox authorization as /api/v1/webhooks/endpoints. A repeated idempotency_key with the same input replays the stored result.

    Annotations the server declares

    • title: Test Webhook Endpoint With Sample
    • readOnlyHint: false
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: true
  • update_webhookConfigure webhook endpoint URL and event subscriptions for real-time notifications.Can send data outNo findings

    Why it is in this group

    • The name "update_webhook" says "update": a word for a change.

    What it can reach

    • outboundThe input takes a URL (input.webhook_url): a request to a URL can carry data to whoever runs it.
    Description, as the server wrote it
    Configure webhook endpoint URL and event subscriptions for real-time notifications. Outbound events are mail.pending_approval, mail.submitted, mail.ready, mail.mailed, mail.delivered, mail.failed, and mail.cancelled. Canonical physical-custody events are inbound.received and inbound.scan_available. The endpoint needs HTTPS and a 2xx response within 10 seconds.

    Annotations the server declares

    • title: Update Webhook
    • readOnlyHint: false
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
Only reads 30Show 30 read-only tools
  • get_agent_inbox_activityDeprecated: superseded by the tool named in _meta.successor.Reads private dataReads text others write2 findings

    Why it is in this group

    • The name "get_agent_inbox_activity" says "get": a word for reading.

    What it can reach

    • private-dataThe name "get_agent_inbox_activity" says "get" and "inbox": it reads mail.
    • untrusted-inputThe name "get_agent_inbox_activity" says "get" and "inbox": it reads inbound email, which other people can write.

    Findings

    • Info

      The description mentions "credential".

      Matched text

      description: "...s, current member approval, assignment, credential environment and SQL authorization are s..."

      What this check looks for
    • Warn

      The description names a tool "in" that this server does not list.

      Matched text

      description: "Deprecated: superseded by the tool named in _meta.successor. Still served unchanged..."

      What this check looks for
    Description, as the server wrote it
    Deprecated: superseded by the tool named in _meta.successor. Still served unchanged during the alias window. Read current authorized inbox retrieval receipts and agent-reported outcomes through /v1/agent-inbox/activity. Optional item_id filters one item; needs agent.inbox.read. Sample activity also needs agent.read. A receipt is not proof a worker followed its duties or that staff completed handling. Required scopes: agent.inbox.read. Agent-scoped bearer key required; member/facility keys cannot impersonate an agent. Rollout switches, current member approval, assignment, credential environment and SQL authorization are shared with REST.

    Annotations the server declares

    • title: Get Agent Inbox Activity
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_agent_inbox_contextDeprecated: superseded by the tool named in _meta.successor.Reads private dataReads text others write2 findings

    Why it is in this group

    • The name "get_agent_inbox_context" says "get": a word for reading.

    What it can reach

    • private-dataThe name "get_agent_inbox_context" says "get" and "inbox": it reads mail.
    • untrusted-inputThe name "get_agent_inbox_context" says "get" and "inbox": it reads inbound email, which other people can write.

    Findings

    • Info

      The description mentions "credential".

      Matched text

      description: "...s, current member approval, assignment, credential environment and SQL authorization are s..."

      What this check looks for
    • Warn

      The description names a tool "in" that this server does not list.

      Matched text

      description: "Deprecated: superseded by the tool named in _meta.successor. Still served unchanged..."

      What this check looks for
    Description, as the server wrote it
    Deprecated: superseded by the tool named in _meta.successor. Still served unchanged during the alias window. Versioned context/reporting reader through /v1/agent-inbox/:id/context, not all-source history. Member-sample context is the original envelope/letter, not later facility captures. Returns versioned duties, ordered OCR and opaque context_version; only the returned context can be acknowledged. Records retrieval, not processing or facility completion. OCR text comes from the mail piece; missing, pending and failed text are reported as such. No provider call or signed image URL. Required scopes: agent.inbox.read, agent.read. Agent-scoped bearer key required; member/facility keys cannot impersonate an agent. Rollout switches, current member approval, assignment, credential environment and SQL authorization are shared with REST.

    Annotations the server declares

    • title: Get Agent Inbox Context
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_agent_inbox_handlingDeprecated: superseded by the tool named in _meta.successor.Reads private dataReads text others write2 findings

    Why it is in this group

    • The name "get_agent_inbox_handling" says "get": a word for reading.

    What it can reach

    • private-dataThe name "get_agent_inbox_handling" says "get" and "inbox": it reads mail.
    • untrusted-inputThe name "get_agent_inbox_handling" says "get" and "inbox": it reads inbound email, which other people can write.

    Findings

    • Info

      The description mentions "credential".

      Matched text

      description: "...s, current member approval, assignment, credential environment and SQL authorization are s..."

      What this check looks for
    • Warn

      The description names a tool "in" that this server does not list.

      Matched text

      description: "Deprecated: superseded by the tool named in _meta.successor. Still served unchanged..."

      What this check looks for
    Description, as the server wrote it
    Deprecated: superseded by the tool named in _meta.successor. Still served unchanged during the alias window. Read /v1/agent-inbox/:id/handling capabilities, version and history. Gated private-mail Live and member-sample Sandbox workflow only; no generated fixtures or managed-PMB fallback. Server chooses source and facility. Handling version is distinct from OCR context_version. Vault is unavailable; physical scans retain the existing owner-confirmed workflow. Required scopes: agent.inbox.read, agent.read. Agent-scoped bearer key required; member/facility keys cannot impersonate an agent. Rollout switches, current member approval, assignment, credential environment and SQL authorization are shared with REST.

    Annotations the server declares

    • title: Get Agent Inbox Handling
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_agent_inbox_scanDeprecated: superseded by the tool named in _meta.successor.Reads private dataReads text others write2 findings

    Why it is in this group

    • The name "get_agent_inbox_scan" says "get": a word for reading.

    What it can reach

    • private-dataThe name "get_agent_inbox_scan" says "get" and "inbox": it reads mail.
    • untrusted-inputThe name "get_agent_inbox_scan" says "get" and "inbox": it reads inbound email, which other people can write.

    Findings

    • Info

      The description mentions "credential".

      Matched text

      description: "...s, current member approval, assignment, credential environment and SQL authorization are s..."

      What this check looks for
    • Warn

      The description names a tool "in" that this server does not list.

      Matched text

      description: "Deprecated: superseded by the tool named in _meta.successor. Still served unchanged..."

      What this check looks for
    Description, as the server wrote it
    Deprecated: superseded by the tool named in _meta.successor. Still served unchanged during the alias window. Read one exact completed facility capture and its saved page OCR through /v1/agent-inbox/:id/scans/:operationId. Sandbox agent key assigned to the member sample only. operationId is the capture's request_id, not a page or bundle ID. Returns scan/ocr page IDs, order, source_sha256 and OCR status (null, queued, processing, ready, needs_review or failed). These captures are separate from seeded sample /context and have no context_version, so they cannot be acknowledged as an outcome. No image URLs, OCR trigger, retry, receipt, email or charge. Required scopes: agent.inbox.read, agent.read. Agent-scoped bearer key required; member/facility keys cannot impersonate an agent. Rollout switches, current member approval, assignment, credential environment and SQL authorization are shared with REST.

    Annotations the server declares

    • title: Get Agent Inbox Scan
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_agent_inbox_sourcesDeprecated: superseded by the tool named in _meta.successor.Reads private dataReads text others write2 findings

    Why it is in this group

    • The name "get_agent_inbox_sources" says "get": a word for reading.

    What it can reach

    • private-dataThe name "get_agent_inbox_sources" says "get" and "inbox": it reads mail.
    • untrusted-inputThe name "get_agent_inbox_sources" says "get" and "inbox": it reads inbound email, which other people can write.

    Findings

    • Info

      The description mentions "credential".

      Matched text

      description: "...s, current member approval, assignment, credential environment and SQL authorization are s..."

      What this check looks for
    • Warn

      The description names a tool "in" that this server does not list.

      Matched text

      description: "Deprecated: superseded by the tool named in _meta.successor. Still served unchanged..."

      What this check looks for
    Description, as the server wrote it
    Deprecated: superseded by the tool named in _meta.successor. Still served unchanged during the alias window. Read one assigned item's saved page OCR through GET /v1/agent-inbox/:id/sources: exterior, requested-content and original sample_letter pages, each with page_id, request_id/bundle_id, source_sha256 and review flags. Requires assigned-inbox and indexed-source reads enabled for the account, plus agent.inbox.read + agent.read. Field meanings and sort order are in outputSchema. Pages: limit 1–50 (default 25), offset 0–10000; has_more marks more pages. Rate limited; a 429 may carry Retry-After. Creates no acknowledgment version, extraction, receipt, provider call, signed image URL, email or charge. Managed-PMB custody and forwarded email remain separate and excluded. Required scopes: agent.inbox.read, agent.read. Agent-scoped bearer key required; member/facility keys cannot impersonate an agent. Rollout switches, current member approval, assignment, credential environment and SQL authorization are shared with REST.

    Annotations the server declares

    • title: Get Agent Inbox Sources
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_batch_mailA batch with its status (draft → paid → generating → submitted → in_progress → completed | partially_completed | failed; cancelled), recipient summary, pricing (credits_charged, re…Reads private dataReads text others writeNo findings

    Why it is in this group

    • The name "get_batch_mail" says "get": a word for reading.

    What it can reach

    • private-dataThe name "get_batch_mail" says "get" and "mail": it reads mail.
    • untrusted-inputThe name "get_batch_mail" says "get" and "mail": it reads inbound email, which other people can write.
    Description, as the server wrote it
    A batch with its status (draft → paid → generating → submitted → in_progress → completed | partially_completed | failed; cancelled), recipient summary, pricing (credits_charged, refunded_cents), field_kit, tracking (QR pieces scanned and confirmed), links and the first 50 pieces, each with its qr (code, link, your ref, scans, first_scanned_at, confirmed_at, expires_at; sandbox verify pieces also carry short_code). Poll it after confirm; batch.submitted / batch.completed also arrive on the agent's webhook. Every piece: page with offset and limit, or ask for the pieces scanned, confirmed (verify: the address verified) or not_scanned with qr; REST: GET /v1/batch-mail/{id}/pieces.

    Annotations the server declares

    • title: Get Batch Mail
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_batch_mail_proofA proof for the person checking the batch: a one-hour link to a PDF of the template as recipient N's piece prints (their fields filled in, the QR code where it goes, a sample code …Reads private dataReads text others writeNo findings

    Why it is in this group

    • The name "get_batch_mail_proof" says "get": a word for reading.

    What it can reach

    • private-dataThe name "get_batch_mail_proof" says "get" and "mail": it reads mail.
    • untrusted-inputThe name "get_batch_mail_proof" says "get" and "mail": it reads inbound email, which other people can write.
    Description, as the server wrote it
    A proof for the person checking the batch: a one-hour link to a PDF of the template as recipient N's piece prints (their fields filled in, the QR code where it goes, a sample code until the pieces exist, and the insert), then the address label the facility applies at its real size, next to the return address. Also returns the label's lines as text. proof_url opens in any browser or IDE for an hour, with no key, so the person checking the batch can see it before the batch is confirmed. (links.proof on the batch is the API route itself and needs the API key.)

    Annotations the server declares

    • title: Get Batch Mail Proof
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_facility_messagesRead the message thread with your mailbox facility.Reads private dataReads text others writeNo findings

    Why it is in this group

    • The name "get_facility_messages" says "get": a word for reading.

    What it can reach

    • private-dataThe name "get_facility_messages" says "get" and "message": it reads messages.
    • private-dataThe description's first sentence says "read" and "message": it reads messages.
    • untrusted-inputThe name "get_facility_messages" says "get" and "message": it reads messages, which other people can write.
    • untrusted-inputThe description's first sentence says "read" and "message": it reads messages, which other people can write.
    Description, as the server wrote it
    Read the message thread with your mailbox facility. Facility routing is automatic. Returns messages in reverse chronological order with sender role (member, facility, agent). Supports cursor-based pagination. Automatically marks facility messages as read.

    Annotations the server declares

    • title: Get Facility Messages
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_inbound_activityRead the inbound timeline through GET /v1/inbound-activity: received, action proposed/requested/started/completed/rejected events for this key's visible items, newest first.No findings

    Why it is in this group

    • The name "get_inbound_activity" says "get": a word for reading.
    Description, as the server wrote it
    Read the inbound timeline through GET /v1/inbound-activity: received, action proposed/requested/started/completed/rejected events for this key's visible items, newest first. Optional item_id narrows to one item. A row proves the event was recorded, not OCR accuracy or physical completion. Key scope: inbound.item.read (agent.inbox.read also works). Member keys read their own live items; agent keys read items bound or assigned to them, in their key's environment. Proposals are agent-only and always wait for the owner: they never charge credits, buy postage, open, forward or discard anything by themselves. Forward destinations and discard proposals are checked against the agent's structured inbound policy. Image URLs are never issued to agents.

    Annotations the server declares

    • title: Get Inbound Activity
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_inbound_itemRead one inbound item through GET /v1/inbound-items/:id: canonical status, sender (staff-entered or exterior OCR guess), mailbox PMB, assigned agent, every saved page (exterior, in…Reads private dataReads text others writeNo findings

    Why it is in this group

    • The name "get_inbound_item" says "get": a word for reading.

    What it can reach

    • private-dataThe description's first sentence says "read" and "mailbox": it reads mail.
    • untrusted-inputThe description's first sentence says "read" and "mailbox": it reads inbound email, which other people can write.
    Description, as the server wrote it
    Read one inbound item through GET /v1/inbound-items/:id: canonical status, sender (staff-entered or exterior OCR guess), mailbox PMB, assigned agent, every saved page (exterior, interior, evidence) with OCR status and text when ready, the full action history, and current scan/forward quotes. signed_urls: true adds one-hour page image URLs on live keys (sandbox keys get url: null); page text is OCR of the mail itself. Key scope: inbound.item.read (agent.inbox.read also works). Member keys read their own live items; agent keys read items bound or assigned to them, in their key's environment. Proposals are agent-only and always wait for the owner: they never charge credits, buy postage, open, forward or discard anything by themselves. Forward destinations and discard proposals are checked against the agent's structured inbound policy. Image URLs are never issued to agents.

    Annotations the server declares

    • title: Get Inbound Item
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_inbound_item_sourcesDeprecated: superseded by the tool named in _meta.successor.1 finding

    Why it is in this group

    • The name "get_inbound_item_sources" says "get": a word for reading.

    Findings

    • Warn

      The description names a tool "in" that this server does not list.

      Matched text

      description: "Deprecated: superseded by the tool named in _meta.successor. Still served unchanged..."

      What this check looks for
    Description, as the server wrote it
    Deprecated: superseded by the tool named in _meta.successor. Still served unchanged during the alias window. Read saved exterior and authorized completed inside-page evidence through GET /v1/inbound-items/:id/sources. Results are grouped by item.id/reference_code and owner mailbox, then request_id/bundle_id, then page_number (up to 100). owner identifies the member/renter, assigned agent, facility, canonical mailbox record/public ID and PMB allocation/number. PMB numbers are facility-scoped and come from verified mailbox records, not OCR. Each page carries page_id, source_sha256, status and uncertain_spans. Missing, queued, failed, blank and needs_review are distinct; scan completion does not prove readable text. Reads have no context_version or acknowledgment authority. Saved-data reads only: no extraction, retry, signed image URL, facility action, provider call, email or charge. Exterior automation and authorized inside-page extraction have independent activation that these readers leave unchanged. Limit 1–50 (default 25), offset 0–10000; pagination.has_more marks more results. Results are not snapshot-consistent. Rate limited; a 429 may carry Retry-After. No new exterior-ready webhook or unbounded export is promised. Requires saved-document reads enabled for the account, an active eligible managed mailbox, and both inbound.item.read and inbound.item.scan.read, even for empty searches. Live member keys are owner-scoped; Live agent keys see only their current assigned mail. No facility, Sandbox or package.read compatibility access. A member key cannot select agent duties.

    Annotations the server declares

    • title: Get Inbound Item Sources
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_inbound_mailGet one forwarded inbound mail item with compact draft_context by default.Reads private dataReads text others writeNo findings

    Why it is in this group

    • The name "get_inbound_mail" says "get": a word for reading.

    What it can reach

    • private-dataThe name "get_inbound_mail" says "get" and "mail": it reads mail.
    • private-dataThe description's first sentence says "get" and "mail": it reads mail.
    • untrusted-inputThe name "get_inbound_mail" says "get" and "mail": it reads inbound email, which other people can write.
    • untrusted-inputThe description's first sentence says "get" and "mail": it reads inbound email, which other people can write.
    Description, as the server wrote it
    Get one forwarded inbound mail item with compact draft_context by default. Use this before drafting an outbound reply when you need sender context, reply contact candidates, deadline clues, source files, and thread linkage in one stable payload.

    Annotations the server declares

    • title: Get Inbound Mail
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_inbound_pagesRead an item's scanned pages through GET /v1/inbound-items/:id/pages: exterior (the envelope, present from arrival) and interior pages (after a completed Open & scan), each with oc…Reads text others writeNo findings

    Why it is in this group

    • The name "get_inbound_pages" says "get": a word for reading.

    What it can reach

    • untrusted-inputThe description's first sentence says "read" and "review": it reads comments, which other people can write.
    Description, as the server wrote it
    Read an item's scanned pages through GET /v1/inbound-items/:id/pages: exterior (the envelope, present from arrival) and interior pages (after a completed Open & scan), each with ocr_status, the page's OCR text when ready or needs_review, is_blank and uncertain_spans. signed_urls: true adds one-hour image URLs on live keys (sandbox keys always get url: null). Key scope: inbound.item.read (agent.inbox.read also works). Member keys read their own live items; agent keys read items bound or assigned to them, in their key's environment. Proposals are agent-only and always wait for the owner: they never charge credits, buy postage, open, forward or discard anything by themselves. Forward destinations and discard proposals are checked against the agent's structured inbound policy. Image URLs are never issued to agents.

    Annotations the server declares

    • title: Get Inbound Pages
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_mailboxGet your agent's real mailing address: street address + mailbox number (PMB) at the Manhattan Beach, CA facility.Reads private dataReads text others writeNo findings

    Why it is in this group

    • The name "get_mailbox" says "get": a word for reading.

    What it can reach

    • private-dataThe name "get_mailbox" says "get" and "mailbox": it reads mail.
    • private-dataThe description's first sentence says "get" and "mailbox": it reads mail.
    • untrusted-inputThe name "get_mailbox" says "get" and "mailbox": it reads inbound email, which other people can write.
    • untrusted-inputThe description's first sentence says "get" and "mailbox": it reads inbound email, which other people can write.
    Description, as the server wrote it
    Get your agent's real mailing address: street address + mailbox number (PMB) at the Manhattan Beach, CA facility. The PMB number is assigned after USPS Form 1583 verification. Inbound context from addresses the operator already uses arrives through a separate private intake alias for scans, PDFs, photos, provider notices, and notes.

    Annotations the server declares

    • title: Get Mailbox
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_mailbox_mdReturns this agent's MAILBOX.md, the renter's document of mail-handling preferences: the full document text, version number, content hash, and last update timestamp.Reads private dataReads text others writeNo findings

    Why it is in this group

    • The name "get_mailbox_md" says "get": a word for reading.

    What it can reach

    • private-dataThe name "get_mailbox_md" says "get" and "mailbox": it reads mail.
    • private-dataThe description's first sentence says "return" and "mailbox": it reads mail.
    • untrusted-inputThe name "get_mailbox_md" says "get" and "mailbox": it reads inbound email, which other people can write.
    • untrusted-inputThe description's first sentence says "return" and "mailbox": it reads inbound email, which other people can write.
    Description, as the server wrote it
    Returns this agent's MAILBOX.md, the renter's document of mail-handling preferences: the full document text, version number, content hash, and last update timestamp. Its version number is the mailbox_md_version that outbound sends carry for sync verification.

    Annotations the server declares

    • title: Get MAILBOX.md
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_outbound_mailGet full details of an outbound mail job including recipient address, mail class, page count, cost breakdown, current status, failure metadata, document metadata, and fulfillment p…Reads private dataReads text others writeNo findings

    Why it is in this group

    • The name "get_outbound_mail" says "get": a word for reading.

    What it can reach

    • private-dataThe name "get_outbound_mail" says "get" and "mail": it reads mail.
    • private-dataThe description's first sentence says "get" and "mail": it reads mail.
    • untrusted-inputThe name "get_outbound_mail" says "get" and "mail": it reads inbound email, which other people can write.
    • untrusted-inputThe description's first sentence says "get" and "mail": it reads inbound email, which other people can write.
    Description, as the server wrote it
    Get full details of an outbound mail job including recipient address, mail class, page count, cost breakdown, current status, failure metadata, document metadata, and fulfillment photos. Legacy plaintext records may include direct document URLs; encrypted source documents are retrieved through the REST document endpoint with document.read scope.

    Annotations the server declares

    • title: Get Outbound Mail
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_postal_threadGet one physical-mail thread with optional timeline events.Reads private dataReads text others writeNo findings

    Why it is in this group

    • The name "get_postal_thread" says "get": a word for reading.

    What it can reach

    • private-dataThe name "get_postal_thread" says "get" and "thread": it reads messages.
    • private-dataThe description's first sentence says "get" and "mail": it reads mail.
    • untrusted-inputThe name "get_postal_thread" says "get" and "thread": it reads messages, which other people can write.
    • untrusted-inputThe description's first sentence says "get" and "mail": it reads inbound email, which other people can write.
    Description, as the server wrote it
    Get one physical-mail thread with optional timeline events. Use this to explain how a generated outbound mail piece relates back to prior inbound scans and review decisions.

    Annotations the server declares

    • title: Get Postal Thread
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • get_usageGet usage summary, billing events, and prepaid credit balance for a time period.No findings

    Why it is in this group

    • The name "get_usage" says "get": a word for reading.
    Description, as the server wrote it
    Get usage summary, billing events, and prepaid credit balance for a time period. Returns itemized events (scans, forwards, mail sends) with costs, period totals, and credits. Defaults to the current billing period if no dates are specified. Use this in Cursor/MCP chat when the human asks how many mailbox.bot credits are left; answer with the prepaid balance and explain that only the signed-in human can add funds.

    Annotations the server declares

    • title: Get Usage
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • list_agent_inboxDeprecated: superseded by the tool named in _meta.successor.Reads private dataReads text others write2 findings

    Why it is in this group

    • The name "list_agent_inbox" says "list": a word for reading.

    What it can reach

    • private-dataThe name "list_agent_inbox" says "list" and "inbox": it reads mail.
    • untrusted-inputThe name "list_agent_inbox" says "list" and "inbox": it reads inbound email, which other people can write.

    Findings

    • Info

      The description mentions "credential".

      Matched text

      description: "...s, current member approval, assignment, credential environment and SQL authorization are s..."

      What this check looks for
    • Warn

      The description names a tool "in" that this server does not list.

      Matched text

      description: "Deprecated: superseded by the tool named in _meta.successor. Still served unchanged..."

      What this check looks for
    Description, as the server wrote it
    Deprecated: superseded by the tool named in _meta.successor. Still served unchanged during the alias window. List/search current assigned postal mail using the same /v1/agent-inbox service. q is literal AND search of IDs, stored metadata and authorized stored OCR (NFKC/lowercase, max 200 characters and 12 terms), before pagination; nonempty q additionally needs agent.read. When indexed-source reads are enabled for the account, indexed search includes saved exterior and requested-content pages, including later sample captures. Optional received_after (inclusive) and received_before (exclusive) require that rollout and both read scopes; they filter the stored intake timestamp, not a date printed on a letter. Sender and address keywords match through q and OCR text; there is no separate sender field. Returns metadata, not text. No fuzzy search or new OCR. Sandbox sees isolated fixtures and enabled assigned member samples; Live sees approved assigned private mail. Pagination uses pagination.next_offset up to offset 10000. Required scopes: agent.inbox.read. Agent-scoped bearer key required; member/facility keys cannot impersonate an agent. Rollout switches, current member approval, assignment, credential environment and SQL authorization are shared with REST.

    Annotations the server declares

    • title: List Agent Inbox
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • list_agent_inbox_scansDeprecated: superseded by the tool named in _meta.successor.Reads private dataReads text others write2 findings

    Why it is in this group

    • The name "list_agent_inbox_scans" says "list": a word for reading.

    What it can reach

    • private-dataThe name "list_agent_inbox_scans" says "list" and "inbox": it reads mail.
    • untrusted-inputThe name "list_agent_inbox_scans" says "list" and "inbox": it reads inbound email, which other people can write.

    Findings

    • Info

      The description mentions "credential".

      Matched text

      description: "...s, current member approval, assignment, credential environment and SQL authorization are s..."

      What this check looks for
    • Warn

      The description names a tool "in" that this server does not list.

      Matched text

      description: "Deprecated: superseded by the tool named in _meta.successor. Still served unchanged..."

      What this check looks for
    Description, as the server wrote it
    Deprecated: superseded by the tool named in _meta.successor. Still served unchanged during the alias window. List/search completed facility-captured scan jobs for this assigned member sample through /v1/agent-inbox/:id/scans. Sandbox agent key only, with both read scopes even without q. q is literal AND search of saved capture identifiers and authorized captured-page OCR, normalized NFKC/lowercase, max 200 characters and 12 terms, before pagination. Returns request_id, bundle_id, page count and OCR status, not text. Pagination uses pagination.has_more with offset plus scans.length, up to offset 10000. A completed scan may have null, queued, processing, ready, needs_review or failed OCR; completion is not extraction success. Separate from seeded sample /context; after indexed rollout, item search also includes these captures. No provider, retry, acknowledgment or receipt is triggered. Required scopes: agent.inbox.read, agent.read. Agent-scoped bearer key required; member/facility keys cannot impersonate an agent. Rollout switches, current member approval, assignment, credential environment and SQL authorization are shared with REST.

    Annotations the server declares

    • title: List Agent Inbox Scans
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • list_batch_mailThe caller's batches, newest first.Reads private dataReads text others writeNo findings

    Why it is in this group

    • The name "list_batch_mail" says "list": a word for reading.

    What it can reach

    • private-dataThe name "list_batch_mail" says "list" and "mail": it reads mail.
    • untrusted-inputThe name "list_batch_mail" says "list" and "mail": it reads inbound email, which other people can write.
    Description, as the server wrote it
    The caller's batches, newest first. Sandbox keys list sandbox batches only; agent keys see their own agent's batches.

    Annotations the server declares

    • title: List Batch Mail
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • list_facility_conversationsList your conversation with your mailbox facility, including its unread message count and last message preview.Reads private dataReads text others writeNo findings

    Why it is in this group

    • The name "list_facility_conversations" says "list": a word for reading.

    What it can reach

    • private-dataThe name "list_facility_conversations" says "list" and "conversation": it reads messages.
    • private-dataThe description's first sentence says "list" and "conversation": it reads messages.
    • untrusted-inputThe name "list_facility_conversations" says "list" and "conversation": it reads messages, which other people can write.
    • untrusted-inputThe description's first sentence says "list" and "conversation": it reads messages, which other people can write.
    Description, as the server wrote it
    List your conversation with your mailbox facility, including its unread message count and last message preview. Facility routing is automatic.

    Annotations the server declares

    • title: List Conversations
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • list_inbound_forwarding_addressesList the renter’s private inbound forwarding aliases on forward.mailbox.bot.Reads private dataReads text others writeNo findings

    Why it is in this group

    • The name "list_inbound_forwarding_addresses" says "list": a word for reading.

    What it can reach

    • private-dataThe description's first sentence says "list" and "mailbox": it reads mail.
    • untrusted-inputThe description's first sentence says "list" and "mailbox": it reads inbound email, which other people can write.
    Description, as the server wrote it
    List the renter’s private inbound forwarding aliases on forward.mailbox.bot. These are the unique intake email addresses an operator, assistant, provider, or external agent can forward scans, PDFs, photos, provider notices, notes, and other context-aware documents to so mailbox.bot can build OCR-backed inbound context. Forwarding/emailing attachments here initiates OCR/extraction; this tool discovers the address and does not upload files directly into OCR. The alias is member-scoped, so live and sandbox agent keys for the same member resolve to the same intake address.

    Annotations the server declares

    • title: List Inbound Forwarding Addresses
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • list_inbound_itemsDeprecated: superseded by the tool named in _meta.successor.1 finding

    Why it is in this group

    • The name "list_inbound_items" says "list": a word for reading.

    Findings

    • Warn

      The description names a tool "in" that this server does not list.

      Matched text

      description: "Deprecated: superseded by the tool named in _meta.successor. Still served unchanged..."

      What this check looks for
    Description, as the server wrote it
    Deprecated: superseded by the tool named in _meta.successor. Still served unchanged during the alias window. Search managed-renter physical mail through GET /v1/inbound-items?search_mode=documents. q uses literal AND terms across authorized saved metadata, exterior OCR and completed requested inside-page OCR (NFKC/lowercase, max 200 characters and 12 terms). Terms may match different pages of one item. received_after is inclusive and received_before exclusive; these filter intake time, not a printed letter date. status and kind narrow canonical item metadata. Sender and address terms match through q; there are no separate sender fields. Returns item metadata and canonical owner/mailbox identity, not page text. Saved-data reads only: no extraction, retry, signed image URL, facility action, provider call, email or charge. Exterior automation and authorized inside-page extraction have independent activation that these readers leave unchanged. Limit 1–50 (default 25), offset 0–10000; pagination.has_more marks more results. Results are not snapshot-consistent. Rate limited; a 429 may carry Retry-After. No new exterior-ready webhook or unbounded export is promised. Requires saved-document reads enabled for the account, an active eligible managed mailbox, and both inbound.item.read and inbound.item.scan.read, even for empty searches. Live member keys are owner-scoped; Live agent keys see only their current assigned mail. No facility, Sandbox or package.read compatibility access. A member key cannot select agent duties.

    Annotations the server declares

    • title: List Inbound Items
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • list_inbound_mailList forwarded inbound mail items captured from private forwarding aliases.Reads private dataReads text others writeNo findings

    Why it is in this group

    • The name "list_inbound_mail" says "list": a word for reading.

    What it can reach

    • private-dataThe name "list_inbound_mail" says "list" and "mail": it reads mail.
    • private-dataThe description's first sentence says "list" and "mail": it reads mail.
    • untrusted-inputThe name "list_inbound_mail" says "list" and "mail": it reads inbound email, which other people can write.
    • untrusted-inputThe description's first sentence says "list" and "mail": it reads inbound email, which other people can write.
    Description, as the server wrote it
    List forwarded inbound mail items captured from private forwarding aliases. Default output includes compact draft_context so an LLM or external agent can reason about OCR context, reply contact candidates, deadlines, and thread linkage before generating outbound mail.

    Annotations the server declares

    • title: List Inbound Mail
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • list_outbound_mailList outbound mail jobs with status tracking.Reads private dataReads text others writeNo findings

    Why it is in this group

    • The name "list_outbound_mail" says "list": a word for reading.

    What it can reach

    • private-dataThe name "list_outbound_mail" says "list" and "mail": it reads mail.
    • private-dataThe description's first sentence says "list" and "mail": it reads mail.
    • untrusted-inputThe name "list_outbound_mail" says "list" and "mail": it reads inbound email, which other people can write.
    • untrusted-inputThe description's first sentence says "list" and "mail": it reads inbound email, which other people can write.
    Description, as the server wrote it
    List outbound mail jobs with status tracking. Returns mail ID, recipient, mail class, status, cost, timestamps, and failure metadata. Filter by status, created_at date range, or search recipient/address/tracking/agent notes.

    Annotations the server declares

    • title: List Outbound Mail
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • list_postal_threadsList physical-mail threads that group inbound mail context, human review, and outbound sends.Reads private dataReads text others writeNo findings

    Why it is in this group

    • The name "list_postal_threads" says "list": a word for reading.

    What it can reach

    • private-dataThe name "list_postal_threads" says "list" and "thread": it reads messages.
    • private-dataThe description's first sentence says "list" and "mail": it reads mail.
    • untrusted-inputThe name "list_postal_threads" says "list" and "thread": it reads messages, which other people can write.
    • untrusted-inputThe description's first sentence says "list" and "mail": it reads inbound email, which other people can write.
    Description, as the server wrote it
    List physical-mail threads that group inbound mail context, human review, and outbound sends. Use this to understand which inbound items and outbound documents belong to the same business workflow.

    Annotations the server declares

    • title: List Postal Threads
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • list_webhook_deliveriesRead the latest 50 deliveries and their exact JSON payloads (identifiers, sender line, dates, matched terms; never page text) for one webhook.No findings

    Why it is in this group

    • The name "list_webhook_deliveries" says "list": a word for reading.
    Description, as the server wrote it
    Read the latest 50 deliveries and their exact JSON payloads (identifiers, sender line, dates, matched terms; never page text) for one webhook. HTTP 2xx delivered means receipt, not external-agent processing. No signing secrets or receiver response body. Requires webhook.manage. Uses the same bounded control plane and tenant/agent/Live-Sandbox authorization as /api/v1/webhooks/endpoints. A repeated idempotency_key with the same input replays the stored result.

    Annotations the server declares

    • title: List Webhook Deliveries
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • list_webhook_endpointsList the member's webhooks (name, URL, events, keyword rules, status), without secrets.Reads private dataReads text others writeNo findings

    Why it is in this group

    • The name "list_webhook_endpoints" says "list": a word for reading.

    What it can reach

    • private-dataThe description's first sentence says "list" and "secret": it reads secrets.
    • untrusted-inputThe description's first sentence says "list" and "url": it reads web pages, which anyone can write.
    Description, as the server wrote it
    List the member's webhooks (name, URL, events, keyword rules, status), without secrets. Legacy agent callback settings are separate. Requires webhook.manage. Uses the same bounded control plane and tenant/agent/Live-Sandbox authorization as /api/v1/webhooks/endpoints. A repeated idempotency_key with the same input replays the stored result.

    Annotations the server declares

    • title: List Webhook Endpoints
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false
  • search_inbound_itemsList and search inbound mail through GET /v1/inbound-items, the same list and search the dashboard shows: newest first, with public status, kind (letter | package), the sender line…Reads private dataReads text others writeNo findings

    Why it is in this group

    • The name "search_inbound_items" says "search": a word for reading.

    What it can reach

    • private-dataThe description's first sentence says "list" and "mail": it reads mail.
    • untrusted-inputThe description's first sentence says "list" and "mail": it reads inbound email, which other people can write.
    Description, as the server wrote it
    List and search inbound mail through GET /v1/inbound-items, the same list and search the dashboard shows: newest first, with public status, kind (letter | package), the sender line as read from the envelope, the envelope page and any open action. q is a literal AND search (max 12 terms, 200 characters) over the reference, the sender and the OCR text of every readable page, envelope or contents; status, kind, received_after and received_before (intake time) narrow it. Sandbox keys (sk_agent_test_) see the account's sample letter from Mojave Land Partners (q: "Mojave" finds it); an empty result carries a notice explaining why. Key scope: inbound.item.read (agent.inbox.read also works). Member keys read their own live items; agent keys read items bound or assigned to them, in their key's environment. Proposals are agent-only and always wait for the owner: they never charge credits, buy postage, open, forward or discard anything by themselves. Forward destinations and discard proposals are checked against the agent's structured inbound policy. Image URLs are never issued to agents.

    Annotations the server declares

    • title: Search Inbound Items
    • readOnlyHint: true
    • destructiveHint: false
    • idempotentHint: true
    • openWorldHint: false

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.

6216 paths in this server alone from private data, through text someone else wrote, to a way out. Denying 7 outbound tools breaks every one.

Can send data out7
Can carry data to people or systems: it sends, posts or publishes, or takes a recipient, channel or URL.
deny to break every path
{
  "permissions": {
    "deny": [
      "mcp__mailbox__create_batch_mail",
      "mcp__mailbox__create_test_outbound_mail",
      "mcp__mailbox__create_webhook_endpoint",
      "mcp__mailbox__send_facility_message",
      "mcp__mailbox__send_outbound_mail",
      "mcp__mailbox__send_text_letter",
      "mcp__mailbox__update_webhook"
    ]
  }
}

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.

Tools listed without sign-in

Looked at Nov 20, 2026, 18:31 UTC, unchanged since Oct 29, 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
25
Protocol version
2025-11-25
WWW-Authenticate on 401
not observed
resource_metadata in WWW-Authenticate
not observed
RFC 9728 protected resource metadata
https://mailbox.bot/.well-known/oauth-protected-resource/api/mcp
Authorization servers
https://mailbox.bot
Legacy HTTP+SSE transport
no
Dynamic client registration
yes
Client ID metadata documents
yes
  • HighThe whole server

    The endpoint served tools/list without credentials, and the list has 25 tools that change things (7 destructive, 2 irreversible, 16 write). Whether a call also works without credentials was not tested: tools are never called.

    Matched text

    https://mailbox.bot/api/mcp: tools/list returned 55 tools to an anonymous client, including send_outbound_mail (destructive), cancel_outbound_mail (destructive), update_webhook_endpoint (destructive), rotate_webhook_endpoint_secret (destructive), send_text_letter (destructive) and 20 more

    What this check looks for

Posture changes

  1. Oct 29, 2026
    • Tools that change things, listed without credentials: 20 to 25

Findings

Matches of fixed checks, with the text they matched.

10 warn13 info
Show 17 more findings

History

Changes to its tool definitions and to its remote endpoint, newest first.

  1. Tools41 changed.~41 changedView diff
  2. Tools11 changed.~11 changedView diff
  3. PostureRemote posture changed. Tools that change things, listed without credentials: 20 -> 25.View diff
    • Tools that change things, listed without credentials: 20 to 25
  4. Tools8 added.+8 addedView diff

Install

Pinned to version 1.0.5. Values in angle brackets are yours to fill in.

claude mcp add --transport http mailbox https://mailbox.bot/api/mcp --header 'Authorization: <your-authorization>'
claude mcp add --transport http mailbox https://mailbox.bot/api/mcp --header 'Authorization: <your-authorization>'

Run 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.

.claude/settings.json
{
  "permissions": {
    "allow": [
      "mcp__mailbox__get_mailbox",
      "mcp__mailbox__get_usage",
      "mcp__mailbox__list_inbound_forwarding_addresses",
      "mcp__mailbox__list_inbound_mail",
      "mcp__mailbox__get_inbound_mail",
      "mcp__mailbox__list_postal_threads",
      "mcp__mailbox__get_postal_thread",
      "mcp__mailbox__get_mailbox_md",
      "mcp__mailbox__list_facility_conversations",
      "mcp__mailbox__get_facility_messages",
      "mcp__mailbox__list_outbound_mail",
      "mcp__mailbox__get_outbound_mail",
      "mcp__mailbox__list_agent_inbox",
      "mcp__mailbox__get_agent_inbox_context",
      "mcp__mailbox__get_agent_inbox_sources",
      "mcp__mailbox__list_agent_inbox_scans",
      "mcp__mailbox__get_agent_inbox_scan",
      "mcp__mailbox__get_agent_inbox_activity",
      "mcp__mailbox__get_agent_inbox_handling",
      "mcp__mailbox__list_webhook_endpoints",
      "mcp__mailbox__list_webhook_deliveries",
      "mcp__mailbox__list_inbound_items",
      "mcp__mailbox__get_inbound_item_sources",
      "mcp__mailbox__search_inbound_items",
      "mcp__mailbox__get_inbound_item",
      "mcp__mailbox__get_inbound_pages",
      "mcp__mailbox__get_inbound_activity",
      "mcp__mailbox__get_batch_mail_proof",
      "mcp__mailbox__get_batch_mail",
      "mcp__mailbox__list_batch_mail"
    ],
    "ask": [
      "mcp__mailbox__propose_mailbox_md_edit",
      "mcp__mailbox__send_facility_message",
      "mcp__mailbox__update_webhook",
      "mcp__mailbox__create_test_outbound_mail",
      "mcp__mailbox__advance_test_outbound_mail",
      "mcp__mailbox__report_agent_inbox_outcome",
      "mcp__mailbox__propose_agent_inbox_handling",
      "mcp__mailbox__seed_agent_inbox_sandbox",
      "mcp__mailbox__create_webhook_endpoint",
      "mcp__mailbox__test_webhook_endpoint",
      "mcp__mailbox__test_webhook_endpoint_with_sample",
      "mcp__mailbox__replay_webhook_delivery",
      "mcp__mailbox__quote_inbound_forward",
      "mcp__mailbox__request_inbound_action",
      "mcp__mailbox__quote_outbound_mail",
      "mcp__mailbox__estimate_batch_mail",
      "mcp__mailbox__create_batch_mail",
      "mcp__mailbox__advance_test_batch_mail"
    ],
    "deny": [
      "mcp__mailbox__send_outbound_mail",
      "mcp__mailbox__cancel_outbound_mail",
      "mcp__mailbox__update_webhook_endpoint",
      "mcp__mailbox__rotate_webhook_endpoint_secret",
      "mcp__mailbox__send_text_letter",
      "mcp__mailbox__confirm_batch_mail",
      "mcp__mailbox__cancel_batch_mail"
    ]
  }
}

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 mailbox. Review it before you use it.

Capability tags behind 39 rules
  • mcp__mailbox__get_mailbox private-data, untrusted-input
  • mcp__mailbox__list_inbound_forwarding_addresses private-data, untrusted-input
  • mcp__mailbox__list_inbound_mail private-data, untrusted-input
  • mcp__mailbox__get_inbound_mail private-data, untrusted-input
  • mcp__mailbox__list_postal_threads private-data, untrusted-input
  • mcp__mailbox__get_postal_thread private-data, untrusted-input
  • mcp__mailbox__get_mailbox_md private-data, untrusted-input
  • mcp__mailbox__send_facility_message untrusted-input, outbound
  • mcp__mailbox__list_facility_conversations private-data, untrusted-input
  • mcp__mailbox__get_facility_messages private-data, untrusted-input
  • mcp__mailbox__update_webhook outbound
  • mcp__mailbox__send_outbound_mail untrusted-input, outbound
  • mcp__mailbox__list_outbound_mail private-data, untrusted-input
  • mcp__mailbox__get_outbound_mail private-data, untrusted-input
  • mcp__mailbox__create_test_outbound_mail outbound
  • mcp__mailbox__list_agent_inbox private-data, untrusted-input
  • mcp__mailbox__get_agent_inbox_context private-data, untrusted-input
  • mcp__mailbox__get_agent_inbox_sources private-data, untrusted-input
  • mcp__mailbox__list_agent_inbox_scans private-data, untrusted-input
  • mcp__mailbox__get_agent_inbox_scan private-data, untrusted-input
  • mcp__mailbox__get_agent_inbox_activity private-data, untrusted-input
  • mcp__mailbox__get_agent_inbox_handling private-data, untrusted-input
  • mcp__mailbox__list_webhook_endpoints private-data, untrusted-input
  • mcp__mailbox__create_webhook_endpoint untrusted-input, outbound
  • mcp__mailbox__update_webhook_endpoint untrusted-input
  • mcp__mailbox__test_webhook_endpoint untrusted-input
  • mcp__mailbox__test_webhook_endpoint_with_sample untrusted-input
  • mcp__mailbox__rotate_webhook_endpoint_secret untrusted-input
  • mcp__mailbox__replay_webhook_delivery untrusted-input
  • mcp__mailbox__search_inbound_items private-data, untrusted-input
  • mcp__mailbox__get_inbound_item private-data, untrusted-input
  • mcp__mailbox__get_inbound_pages untrusted-input
  • mcp__mailbox__send_text_letter untrusted-input, outbound
  • mcp__mailbox__create_batch_mail untrusted-input, outbound
  • mcp__mailbox__get_batch_mail_proof private-data, untrusted-input
  • mcp__mailbox__confirm_batch_mail untrusted-input
  • mcp__mailbox__get_batch_mail private-data, untrusted-input
  • mcp__mailbox__list_batch_mail private-data, untrusted-input
  • mcp__mailbox__cancel_batch_mail 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/bot.mailbox/mailbox/verify?hash=4e11abc86ef00c17b8fb79a72a44a550cac27e78a8dc888306362bf14bb361f2" | jq -e '.changed == false'

Pinned hash 4e11abc86ef0, from Nov 20, 2026, 21:18 UTC.

Watch

Get told when its tools change: a new tool, a new description, a new class.

https://radar.anlyon.com/servers/bot.mailbox/mailbox/changes.xml

An Atom feed. Works in any feed reader or Slack's RSS app.

Badge

Anlyon Radar badge for bot.mailbox/mailbox
[![Anlyon Radar](https://radar.anlyon.com/badge/bot.mailbox/mailbox.svg)](https://radar.anlyon.com/servers/bot.mailbox/mailbox)

Source: MCP Registry · repository. Synced Nov 20, 2026, 22:17 UTC.

mailbox.bot MCP server: questions and answers

Answered from the scan above, not from the publisher's description.

Is the mailbox.bot MCP server safe?

No scan can promise that. What its definitions show: 55 tools, 25 of which can change, send or delete something, and 23 findings from fixed checks. Allow the reads, ask before each change, and deny the deletes; the permissions block on this page does that.

Can the mailbox.bot MCP server delete data?

Yes. 7 tools can delete, cancel or revoke something: send_outbound_mail, cancel_outbound_mail, update_webhook_endpoint, rotate_webhook_endpoint_secret, send_text_letter, confirm_batch_mail and 1 more.

Can the mailbox.bot MCP server do things that cannot be undone?

Yes. 2 tools can send, pay, publish or merge: send_facility_message and create_batch_mail.

What tools does the mailbox.bot MCP server have?

55 tools: 7 destructive, 2 irreversible, 16 write, 30 read. send_outbound_mail, cancel_outbound_mail, update_webhook_endpoint, rotate_webhook_endpoint_secret, send_text_letter, confirm_batch_mail, cancel_batch_mail, send_facility_message and 47 more.

Does the mailbox.bot MCP server need sign-in or OAuth?

Its remote endpoint listed its tools to an anonymous client, without sign-in.

Does the mailbox.bot MCP server send data to outside services?

7 tools can carry data out (it sends, posts or publishes, or takes a recipient, channel or URL): send_facility_message, update_webhook, send_outbound_mail, create_test_outbound_mail, create_webhook_endpoint, send_text_letter and 1 more.

Has the mailbox.bot MCP server changed its tools?

Yes, 3 times since Anlyon Radar started tracking it, most recently on Nov 20, 2026. Every change is on this page and in its Atom feed.