Independently Tested & Verified
MCP Directory

Agents don't browse pages — they query the directory directly. Through the REST API or the MCPYP Meta-MCP, your agent finds MCPs the machine way: ranked results, evidence and quality status, and a ready-to-use config.

GET /v1/search — one call, matching MCPs ranked by quality status.

GET https://mcpyp.com/v1/search?need=weather+forecast&region=Canada&language=EN/FR

{
  "results": [
    {
      "name": "Environment Canada Weather",
      "status": "untested",
      "region": "Canada",
      "language": "EN / FR",
      "tools": 4,
      "connect": "https://mcpyp.com/v1/servers/ca-weather"
    }
  ]
}

Meta-MCP — add MCPYP itself as an MCP server, and your agent can search the directory from inside the client:

{
  "mcpServers": {
    "mcpyp": { "url": "https://mcpyp.com/mcp" }
  }
}

Publish

Built an MCP? Get it discovered — and build trust with objective evidence. Here is exactly how a listing goes live:

Submit

Four ways in: repository, server.json, remote URL, or package
One short form

Ownership verification

You prove control of the repo or domain
Disputes go through a takedown path

Security screening

Automated first screen, minutes not days
Basic risk checks before anything is public

Trial listing

Goes live with a Trial badge
A public observation period
We measure it while it runs

Measured evaluation

Deep testing of live behaviour
Graded on measured results only
No badge is guaranteed

Continuous monitoring

Regular re-checks after listing
Evidence and status kept current

Ready to submit?

Independent evidence. Earned trust.

MCP Yellow Pages is an independent testing laboratory for MCP servers. Marketplaces sell you a server. We measure it — and publish what we find.

A laboratory, not a marketplace

We host nothing and sit in no call path. We publish the signpost, never the passage.
A marketplace knows how a product sells. A laboratory knows what it is.
Our revenue comes from the people who use our conclusions — never from the servers we judge.

The record so far

46,933servers in our full census

12,349servers listed and probed

117,500tools measured

162,051checks recorded

98.3%passed our liveness test (12,139 of 12,349)

10security findings on record

What the testing already taught us

HTTP 200 is not alive: Across the full 46,933-server census, 1,949 endpoints answered HTTP but were not live MCP servers. We say so; most directories count them.
22% of publisher repository links in our first deep pass were dead (404).
We publish behaviour, not claims: servers whose live behaviour contradicted their stated auth posture are recorded as measured, not as advertised.

Words, defined

Alive — Passed our protocol-level liveness test. HTTP alone doesn't count.

Deep-tested — Live tool list retrieved and run: 11,293 of 12,349 (91.4%).

Quality — Measured per dimension, shown separately, never blended.

Security — Observable risk evidence. Never a guarantee.

Sponsored — Paid visibility. Always labelled, never a verdict.

Endpoint region — Where the endpoint is served from, as measured.

Global — Served through a worldwide distribution network.

How we test

Every listed server is tested at multiple depths — presence, liveness, live function, quality over time, security — and re-tested continuously. A verdict always carries a date.
We verify what servers return, not just that they respond.
Testing is safe by design: read-only, and never with user tokens.

Security, honestly

Every finding is published as a dated, reproducible evidence report with its limitations stated.
No test can see everything — so we never certify a server "safe". We present risk evidence; the decision stays yours.
Our admission framework is aligned with the OWASP MCP Top 10, the NIST AI Risk Management Framework, MITRE ATLAS, and ISO/IEC 42001.

Independence, in writing

Payment never buys a score, a label, or a rank.
Sponsored placement is labelled and kept out of the measured ranking.
We can publish a harsh verdict on anyone — and we test the marketplaces too.
Wrong measurement? The correction channel is public, and the evidence trail stays visible.

Open method, protected data

Our standards and definitions are open for anyone to audit. Our measured history is earned daily — that is the asset, and it is not for sale.

What this evidence powers — the free directory, pilot pre-integration checks, and what is coming next: Services

Services

What MCPYP sells is independent, measured evidence for decisions about MCP servers.

Available now — free

The directory & search

Free

12,349 MCP servers listed — every one probed for liveness, and 11,293 deep-tested so far, with the evidence published on its profile. Search by what you need, and compare on measured facts, not on claims.

Go to Search

Available as a pilot — delivered manually

Pre-integration check

Pilot

Tell us which MCP server you are about to integrate; we deliver a measured evidence report: what it does, whether it is live and stable, what tools it exposes, and the externally observable risks — so your team decides with evidence, not listings.

During the pilot, every report is prepared and delivered manually — email us and a person takes it from there.

Email [email protected]

Coming soon

Intelligence API

Coming soon

Programmatic access to our measured evidence, by quota — for teams and agents that need the facts inside their own tools.

Enterprise admission & continuous monitoring

Coming soon

An evidence-based admission framework for MCP servers, plus ongoing monitoring — for organisations that run many of them.

Listing in the directory is always free and never for sale. Paid services buy evidence and visibility only — never scores, never verdicts.

Terms of Service

These terms govern your use of the MCP Yellow Pages directory, website, API, and machine tools.

1. What MCPYP is

MCPYP is an independent directory of Model Context Protocol servers with independently measured evidence about each one. Listing in the directory is always free and never for sale. We do not proxy calls, host servers, or store user tokens.

2. Acceptable use

You may use the directory, website, API, and machine tools for lawful purposes, including evaluating MCP servers before you integrate them. You must not use them to harm, disrupt, or mislead others, or to misrepresent MCPYP or its measurements.

3. No bulk scraping or redistribution

Do not bulk-scrape, crawl for republication, or redistribute the catalog or measured data in bulk. Do not circumvent rate limits, paging caps, or other technical measures. We use technical measures, including marker entries, to detect bulk copying. Violations may result in suspension or termination of access and revocation of API keys. We reserve all legal rights.

4. Measured data is evidence, not a guarantee

Our findings are measured evidence at a point in time. MCP servers change constantly, so a finding can go stale. We make no safety guarantees about any server. Verify before you integrate.

5. Publisher content and corrections

Descriptions and claims supplied by publishers belong to the publishers; we label their sources. If you find an error in a listing or a measurement, contact us and we will correct it.

6. Liability

The service is provided as is. To the maximum extent permitted by law, MCPYP is not liable for decisions you make based on directory data or for the behavior of third-party servers.

7. Changes and contact

We may update these terms; continued use after an update means you accept the new terms. Questions: [email protected].

Listing is always free. Paid services buy evidence and visibility only — never scores, never verdicts.