Skip to content

Taste MCP server

app.humantaste/taste-mcphumantaste.appstreamable-httpv2.0.0Scanned

Expert review for AI agents. On-chain proof of human review.

20 tools: 1 do something that can't be undone, 10 change things and 9 only read.

Tools: 0 delete, 1 irreversible, 10 write, 9 read.
  • Reads private data · 0
  • Reads text others write · 6
  • Can send data out · 0
  • No findings

Tools listed without sign-in. Remote posture

With Anlyon

1 tool here does something that can't be undone. Govern it.

  • 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 Taste MCP server can do

The Taste MCP server is published by humantaste.app, a domain the MCP Registry verified. It has 20 tools. It can do things that cannot be undone, such as sending, paying or publishing (request_human_approval) and create or change data (10 tools: request_think_tank_revision, request_illustration_revision, verify_external_source, verify_reasoning_chain and 6 more), but cannot delete data. 9 tools only read.

6 tools bring in text other people can write (list_offerings, prepare_consult_domain_expert, review_content and 3 more).

None of the fixed checks matched anything in its tool definitions. That is not a safety guarantee.

Its remote endpoint lists its tools without sign-in.

Last scanned Nov 10, 2026, version 2.0.0.

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.

20 tools

Can't be undone 1Claude Code: ask
  • request_human_approvalPause your workflow for explicit human approval before executing a high-stakes action.No findings

    Why it is in this group

    • The name "request_human_approval" says "request": a word for a change.
    • The description says "irreversible".
    Description, as the server wrote it
    Pause your workflow for explicit human approval before executing a high-stakes action. Call before any irreversible action — large spend, on-chain transaction, public content publish, customer-facing decision. Returns approved/denied + reasoning. Approvals can be enforced on-chain via the Taste Gatekeeper hook for ACP and ERC-8183 jobs.
Changes things 10Claude Code: ask
  • ask_a_humanAsk a vetted human a plain-text question and get a plain-text answer back, usually within 30 minutes.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
    Ask a vetted human a plain-text question and get a plain-text answer back, usually within 30 minutes. Call when you need a quick human take, gut check, or local/lived-experience answer a model cannot reliably give. No scoring, no verdict, no certificate — just a direct human response. Returns the human's answer, plus an optional note on who answered.
  • consult_domain_expertA vetted human in a named domain answers your question with first-hand knowledge.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
    A vetted human in a named domain answers your question with first-hand knowledge. Pass `domain` to match an expert. Pass `parentSessionId` for follow-ups (cap 3). Returns answer, anonymised role, confidence, first-hand flag. Approved answers get an on-chain Taste cert. Listed domains: musician, cantor, writer, farmer, UX designer, Swedish archipelago resident, Stockholm local, Protestant priest, art curator, museum staff, culture journalist, food critic. Other domains: best-effort 24-48h.
  • order_think_tank_session_30Convene a think tank — a curated team of vetted human experts runs a 30-minute working session on your idea, problem, or decision and returns synthesized directions, ideas with rat…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
    Convene a think tank — a curated team of vetted human experts runs a 30-minute working session on your idea, problem, or decision and returns synthesized directions, ideas with rationale, and recommended next steps. Returns a sessionId + referenceCode immediately — poll get_result with either when the team is done. Includes 1 free revision turn: call request_think_tank_revision with the code to ask for adjustments.
  • order_think_tank_session_60Convene a think tank — a curated team of vetted human experts runs a 60-minute working session on your idea, problem, or decision and returns synthesized directions, ideas with rat…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
    Convene a think tank — a curated team of vetted human experts runs a 60-minute working session on your idea, problem, or decision and returns synthesized directions, ideas with rationale, and recommended next steps. Returns a sessionId + referenceCode immediately — poll get_result with either when the team is done. Includes 2 free revision turns: call request_think_tank_revision with the code to ask for adjustments.
  • request_illustration_revisionSpend a free revision turn on a completed illustration order.No findings

    Why it is in this group

    • The name "request_illustration_revision" says "request": a word for a change.
    Description, as the server wrote it
    Spend a free revision turn on a completed illustration order. Provide the reference code from your original order and describe the adjustments you want — the same illustrator revisits the brief and returns an updated deliverable. Poll get_result with the reference code to retrieve it. Free: no payment required, the turn is included with your order.
  • request_think_tank_revisionSpend a free revision turn on a completed think tank order.No findings

    Why it is in this group

    • The name "request_think_tank_revision" says "request": a word for a change.
    Description, as the server wrote it
    Spend a free revision turn on a completed think tank order. Provide the reference code from your original order and describe the adjustments you want — the same team revisits the brief and returns an updated deliverable. Poll get_result with the reference code to retrieve it. Free: no payment required, the turn is included with your order.
  • review_codeGet a vetted human engineer to review your code, architecture, and design decisions — not just style, but correctness, security, and whether the structure will hold up.Reads text others writeNo findings

    Why it is in this group

    • The description's first verb is "get": a word for reading.
    • The input takes a statement or command to run (input.code), which can change anything it can reach.

    What it can reach

    • untrusted-inputThe name "review_code" says "review": it reads comments, which other people can write.
    • untrusted-inputThe description's first sentence says "get" and "review": it reads comments, which other people can write.
    Description, as the server wrote it
    Get a vetted human engineer to review your code, architecture, and design decisions — not just style, but correctness, security, and whether the structure will hold up. Call before you treat code as done: payment flows, auth, data handling, or any logic where a subtle bug is costly. Pass the code (inline or a publicly accessible URL) and what it is meant to do. Returns verdict (approved / needs_changes / reject), a correctness score, security findings, architecture notes, and suggested changes. Approved code receives a Taste content certificate on-chain.
  • verify_certificateVerify a Taste content certificate on-chain.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
    Verify a Taste content certificate on-chain. Call when consuming content from another source that claims to be human-reviewed — returns reviewer domain, date, offering, verdict, validity, and an evidence tier: "evidence-verified" certs carry the reviewer's EIP-712 verdict signature + their Human Passport score attestation (EAS on Base), independently checkable without trusting Taste. Free. Use to build trust chains: agent A's output is reviewed → certificate issued → agent B consumes content and verifies cert before acting.
  • verify_external_sourceHave a vetted human expert verify whether an external source, project, or claim is trustworthy.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
    Have a vetted human expert verify whether an external source, project, or claim is trustworthy. Call when your output or planned action depends on an external claim you cannot independently verify (crypto project legitimacy, social media authenticity, source credibility, vendor due diligence). Returns verdict, red flags, positive signals, confidence score.
  • verify_reasoning_chainAudit your step-by-step reasoning for mid-chain errors before producing the final answer.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
    Audit your step-by-step reasoning for mid-chain errors before producing the final answer. Call when stakes are high and your chain has 3+ steps — humans catch wrong intermediate steps that final-output checks miss (right-looking answers built on broken intermediate logic). Returns per-step verdict, flagged errors, suggested corrections. Approved chains receive a Taste content certificate on-chain.
Only reads 9Show 9 read-only tools
  • arbitrate_disputeGet an impartial human arbiter to decide whether a deliverable meets an agreed contract, or to settle disagreements between agents.No findings

    Why it is in this group

    • The description's first verb is "get": a word for reading.
    Description, as the server wrote it
    Get an impartial human arbiter to decide whether a deliverable meets an agreed contract, or to settle disagreements between agents. Call when automated evaluation cannot determine quality, when two parties dispute fulfillment, or when subjective judgment is needed for contract assessment. Returns approve/reject verdict, reasoning, deliverable quality rating, contract alignment.
  • get_resultCheck the status of a previously submitted request.No findings

    Why it is in this group

    • The name "get_result" says "get": a word for reading.
    Description, as the server wrote it
    Check the status of a previously submitted request. Returns the full structured deliverable once the expert (or order team) completes the work. Accepts a sessionId or an order reference code (think tank or illustration).
  • list_expert_profilesReturns currently-available expert taglines (pseudonymous descriptions of the kinds of expertise on hand) plus the real-time count of online expert seats and estimated wait.No findings

    Why it is in this group

    • The name "list_expert_profiles" says "list": a word for reading.
    Description, as the server wrote it
    Returns currently-available expert taglines (pseudonymous descriptions of the kinds of expertise on hand) plus the real-time count of online expert seats and estimated wait. Use this as a cheap pre-flight check before calling a paid tool. Taglines describe expertise kinds, not individuals: no per-expert PII is exposed. Free.
  • list_offeringsList all available human expert evaluation offerings with pricing, tool names, and whether each issues an on-chain Taste content certificate.Reads text others writeNo findings

    Why it is in this group

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

    What it can reach

    • untrusted-inputThe description's first sentence says "list" and "issue": it reads issues, which other people can write.
    Description, as the server wrote it
    List all available human expert evaluation offerings with pricing, tool names, and whether each issues an on-chain Taste content certificate. Free.
  • predict_audience_reactionGet a real human matching your target demographic to rate and react to your content as a representative audience member.No findings

    Why it is in this group

    • The description's first verb is "get": a word for reading.
    Description, as the server wrote it
    Get a real human matching your target demographic to rate and react to your content as a representative audience member. Call before A/B test commits, before ad spend, or before distribution decisions. Returns overall rating, criteria scores, qualitative feedback, comparison notes.
  • prepare_consult_domain_expertGet a humantaste.app URL where a human can place a consult_domain_expert order from a browser (Connect MetaMask, pay $15 USDC on Base, session created).Reads text others writeNo findings

    Why it is in this group

    • The description's first verb is "get": a word for reading.

    What it can reach

    • untrusted-inputThe description's first sentence says "get" and "url": it reads web pages, which anyone can write.
    Description, as the server wrote it
    Get a humantaste.app URL where a human can place a consult_domain_expert order from a browser (Connect MetaMask, pay $15 USDC on Base, session created). Use this when your MCP client has no wallet integration (Claude Desktop, generic chat UIs). The URL is pre-filled with the brief you pass in; the user just opens it, reviews, connects a wallet, and pays. Returns the payment URL and the price. Free.
  • prepublish_reviewGet a vetted human expert to review your AI-generated content — text, images, videos, social posts, audio, or any media — for cultural sensitivity, brand safety, derivative risks, …Reads text others writeNo findings

    Why it is in this group

    • The description's first verb is "get": a word for reading.

    What it can reach

    • untrusted-inputThe name "prepublish_review" says "review": it reads comments, which other people can write.
    • untrusted-inputThe description's first sentence says "get" and "review": it reads comments, which other people can write.
    Description, as the server wrote it
    Get a vetted human expert to review your AI-generated content — text, images, videos, social posts, audio, or any media — for cultural sensitivity, brand safety, derivative risks, and audience appeal before you publish. Call before social posts, marketing campaigns, content distribution, or anywhere public-facing. Returns verdict (safe / needs_changes / do_not_publish), flagged issues, fix suggestions. Approved content receives a Taste content certificate on-chain, verifiable downstream.
  • review_contentGet a vetted human expert to review your AI-generated content — text, images, videos, social posts, audio, or any media — for hallucinations, factual errors, weak parts, and domain…Reads text others writeNo findings

    Why it is in this group

    • The description's first verb is "get": a word for reading.

    What it can reach

    • untrusted-inputThe name "review_content" says "review": it reads comments, which other people can write.
    • untrusted-inputThe description's first sentence says "get" and "review": it reads comments, which other people can write.
    Description, as the server wrote it
    Get a vetted human expert to review your AI-generated content — text, images, videos, social posts, audio, or any media — for hallucinations, factual errors, weak parts, and domain mistakes. Call before publishing, before forwarding to another agent, or before acting on the content. Returns verdict, issues found, suggested improvements. Approved outputs receive a Taste content certificate on-chain you can attach as proof of human review.
  • review_planGet a vetted human to review your implementation or project plan before you build.Reads text others writeNo findings

    Why it is in this group

    • The description's first verb is "get": a word for reading.

    What it can reach

    • untrusted-inputThe name "review_plan" says "review": it reads comments, which other people can write.
    • untrusted-inputThe description's first sentence says "get" and "review": it reads comments, which other people can write.
    Description, as the server wrote it
    Get a vetted human to review your implementation or project plan before you build. Call right after you draft a plan for a non-trivial task — a human catches wrong assumptions, missing steps, architectural dead-ends, and risky sequencing while changes are still cheap, before any code is written. Pass the plan, the goal it serves, and any constraints. Returns verdict (proceed / revise / rethink), risks flagged, missing considerations, and sequencing notes. Approved plans receive a Taste content certificate on-chain.

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.

No private-data, untrusted-input, outbound path in this server alone: no tool is tagged private-data or outbound.

Reads private data0
Reads files, mail, messages, databases, secrets, calendars or repositories.
Reads text others write6
Brings in text someone else can write: web pages, inbound email, messages, issues and comments.
Can send data out0
Can carry data to people or systems: it sends, posts or publishes, or takes a recipient, channel or URL.

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, 18:29 UTC, unchanged since Oct 22, 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
11
Protocol version
2025-06-18
WWW-Authenticate on 401
not observed
resource_metadata in WWW-Authenticate
not observed
RFC 9728 protected resource metadata
not observed
Authorization servers
not observed
Legacy HTTP+SSE transport
no
Dynamic client registration
not observed
Client ID metadata documents
not observed
  • HighThe whole server

    The endpoint served tools/list without credentials, and the list has 11 tools that change things (1 irreversible, 10 write). Whether a call also works without credentials was not tested: tools are never called.

    Matched text

    https://api.humantaste.app/mcp/: tools/list returned 20 tools to an anonymous client, including request_human_approval (irreversible), request_think_tank_revision (write), request_illustration_revision (write), verify_external_source (write), verify_reasoning_chain (write) and 6 more

    What this check looks for

Findings

Matches of fixed checks, with the text they matched.

No check matched anything in these tool definitions. That is not a safety guarantee.

History

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

No change seen yet. Its tools are rescanned every day, and a remote endpoint is looked at once a week.

Install

Pinned to version 2.0.0. Values in angle brackets are yours to fill in.

claude mcp add --transport http taste https://api.humantaste.app/mcp/
claude mcp add --transport http taste https://api.humantaste.app/mcp/

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__taste__list_offerings",
      "mcp__taste__list_expert_profiles",
      "mcp__taste__get_result",
      "mcp__taste__prepare_consult_domain_expert",
      "mcp__taste__review_content",
      "mcp__taste__prepublish_review",
      "mcp__taste__predict_audience_reaction",
      "mcp__taste__arbitrate_dispute",
      "mcp__taste__review_plan"
    ],
    "ask": [
      "mcp__taste__request_think_tank_revision",
      "mcp__taste__request_illustration_revision",
      "mcp__taste__verify_external_source",
      "mcp__taste__verify_reasoning_chain",
      "mcp__taste__request_human_approval",
      "mcp__taste__verify_certificate",
      "mcp__taste__order_think_tank_session_30",
      "mcp__taste__order_think_tank_session_60",
      "mcp__taste__consult_domain_expert",
      "mcp__taste__ask_a_human",
      "mcp__taste__review_code"
    ],
    "deny": []
  }
}

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 taste. Review it before you use it.

Capability tags behind 6 rules
  • mcp__taste__list_offerings untrusted-input
  • mcp__taste__prepare_consult_domain_expert untrusted-input
  • mcp__taste__review_content untrusted-input
  • mcp__taste__prepublish_review untrusted-input
  • mcp__taste__review_plan untrusted-input
  • mcp__taste__review_code 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/app.humantaste/taste-mcp/verify?hash=533208d3c9f65f7107dcab83a54856866f927149e1ceeeb277ab9fc670ce92ef" | jq -e '.changed == false'

Pinned hash 533208d3c9f6, from Nov 10, 2026, 06:17 UTC.

Watch

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

https://radar.anlyon.com/servers/app.humantaste/taste-mcp/changes.xml

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

Badge

Anlyon Radar badge for app.humantaste/taste-mcp
[![Anlyon Radar](https://radar.anlyon.com/badge/app.humantaste/taste-mcp.svg)](https://radar.anlyon.com/servers/app.humantaste/taste-mcp)

Source: MCP Registry · repository. Synced Oct 21, 2026, 17:44 UTC.

Taste MCP server: questions and answers

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

Is the Taste MCP server safe?

No scan can promise that. What its definitions show: 20 tools, 11 of which can change, send or delete something, and 0 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 Taste MCP server delete data?

No tool is classed as destructive: going by their definitions, none of them deletes, cancels or revokes anything.

Can the Taste MCP server do things that cannot be undone?

Yes. 1 tool can send, pay, publish or merge: request_human_approval.

What tools does the Taste MCP server have?

20 tools: 1 irreversible, 10 write, 9 read. request_human_approval, request_think_tank_revision, request_illustration_revision, verify_external_source, verify_reasoning_chain, verify_certificate, order_think_tank_session_30, order_think_tank_session_60 and 12 more.

Does the Taste MCP server need sign-in or OAuth?

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

Does the Taste MCP server send data to outside services?

None of its tools sends, posts or publishes, or takes a recipient, channel or URL.

Has the Taste MCP server changed its tools?

Not since Anlyon Radar started tracking it. Last scanned Nov 10, 2026.