# MCP Queen — the evidence layer for MCP

> Search MCP servers by observed evidence, not promises. Operational grades are
> kept separate from Trust Receipts covering security/access boundaries, data
> integrity, citations, advertised claims, response audits, and reviewed field use.

## For agents
- MCP endpoint (streamable HTTP, no auth): https://mcpqueen.com/mcp
  Tools: search_servers (find graded servers by task/category — the discovery
  broker), search_tools (search the actual tool names/descriptions servers
  expose — find a specific capability or data type), list_grades,
  get_server_grade, get_trust_receipt, search_trust_evidence, submit_feedback
  (field reports, quarantined for human review).
- Grades API (JSON, CORS-open): https://mcpqueen.com/api/grades.json
- Per-server Trust Receipt API: https://mcpqueen.com/api/trust/<registry-name>.json
- Direct connection guide for ChatGPT, Claude, Claude Code, and OpenAI agents:
  https://mcpqueen.com/connect
- Founding Scout + Guard pilot for monitored agent stacks:
  https://mcpqueen.com/pilot
- Evidence-led discovery model and machine-readable advantages:
  https://mcpqueen.com/mcp-server-discovery
- Integration guides and runnable examples: https://mcpqueen.com/integrations
  OpenAI, Claude, LangChain, LlamaIndex, Cloudflare Agents, and Hugging Face.
- Architecture and trust boundaries: https://mcpqueen.com/architecture
- Prepared demo chapters, transcript slots, and captions:
  https://mcpqueen.com/demo (footage is not yet published).

## For humans
- Dashboard (sort/filter/search, categories): https://mcpqueen.com/registry
- Plain-language connection guide: https://mcpqueen.com/connect
- Seven-day Scout + Guard stack check: https://mcpqueen.com/pilot
- Why MCP Queen is different from a listing-only directory:
  https://mcpqueen.com/mcp-server-discovery
- Per-server evidence pages: https://mcpqueen.com/s/<registry-name>
- Reviewed real-usage field reports: https://mcpqueen.com/field-reports
- Repository-move guide and dated audit: https://mcpqueen.com/reports/mcp-repository-mobility-2026-08
- Methodology: https://mcpqueen.com/registry#methodology
- Architecture: https://mcpqueen.com/architecture

## For server owners
- Live grade badge: https://mcpqueen.com/badge/<registry-name>.svg
- Grades refresh automatically (~3-day full probe cycle). Fix what the
  evidence shows and the badge updates itself.

Operational grading rubric: reachability 25 / protocol 15 / tooling 35 /
latency 10 / provenance 15. Auth-gated servers are scored on the verifiable
subset and marked provisional. Separately, safe read-only response audits test
call usability, semantic error responses, and whether returned PMID/DOI
identifiers resolve against authoritative sources. Missing trust evidence means
unaudited, never a pass. By the team behind constat.dev and healthai.com.