FaceabotCapability network for AI agents
‹ Back

MCP server reliability data

Faceabot tests public MCP servers around the clock. AI agents can look up the reliability of any server for free, one at a time, and cite it. Here: a sample of the 100 best-proven servers.

Field names and verdicts are in English so any program can read them.

Open data on GitHub

Every Monday, the ranking of the 100 best-proven servers and the global figures are also published as CSV and JSON, with their history, on GitHub: github.com/liametemie-cmd/mcp-reliability. Same terms: cite freely with a link, no bulk copying.

Endpoints (no key, CORS open)
GET https://faceabot.com/api/reliability-feed.json            JSON, the 100 best-proven servers
GET https://faceabot.com/api/reliability-feed.csv             same sample as CSV
GET https://faceabot.com/api/reliability-feed.json?verdict=reliable   only proven-reliable servers
GET https://faceabot.com/api/reliability?url=https://example.com/mcp   one server
Fields

id (official MCP Registry name) · url · verdict · attempts · successes · successRate · consecutiveSuccesses · lastStatus · lastCheckedAt · lastLatencyMs · toolsAdvertised · work (real read-only task: verdict proven_working / partly_working / not_working / testing / not_tested, tool, attempts, successes, sampleInput) · page · badge

Verdicts: reliable ≥ 95 % success over ≥ 3 checks and ≥ 3 in a row · measuring fewer than 3 checks · flaky 60–95 % · failing < 60 % · requires_authentication private server, not a failure.

Terms of use

Free to consult and to cite, with a link to faceabot.com. Bulk copying, scraping, mirroring or redistribution of the Faceabot database is not allowed (EU Directive 96/9/EC, database right). Credit to show when you cite a measurement:

<a href="https://faceabot.com">Reliability data: Faceabot</a>

Or embed the live badge of any server: <img src="https://faceabot.com/badge/<id>.svg">

Method

Each check is an MCP initialize then tools/list; when the server declares a read-only tool (readOnlyHint), Faceabot also calls that one tool with a neutral sample input (e.g. https://example.com, hello world) to check it really returns a result — never a tool that writes, sends, pays or deletes. Everything is sent politely (one request per host at a time, backoff on errors) with the User-Agent FaceabotProbe/1.0 (+https://faceabot.com/reliability; opt-out: https://faceabot.com/api/reliability/opt-out). Server owners can opt out at /api/reliability/opt-out. Publishers can diagnose their own server with the Faceabot doctor.

AI agents: all of this is available without keys over MCP https://faceabot.com/api/mcp · HTTP https://faceabot.com/api/catalog?q=… · Guide : llms.txt

Best tool by topic · Pulse: what is changing in agent commerce · Is your store ready for AI shopping agents? · State of MCP (weekly) · MCP server doctor · Reliability data · Faceabot app · Faceabot → AI assistants · What AI agents cannot find yet · Add Faceabot to your AI · Badge for tool publishers · Help · Privacy · Terms