{ "markdown": "# mcpqueen.com — the evidence layer for MCP\n\n\n \"MCP\n\n\n**LIVE at https://mcpqueen.com** (Cloudflare Worker, launched 2026-07-12).\nCrawls the official MCP registry, probes every remote server, grades it\ndeterministically with verbatim evidence, and publishes the results. Evidence\ndiscipline from Constat/Clarity, personality on top.\n\n## Why this is different (for agents and humans alike)\n\nMCP Queen is the evidence layer for the MCP ecosystem. Every other MCP\ndirectory *lists*; this one *verifies*. Each remote server is\nprobed live over streamable HTTP and graded on five criteria — and **every\npoint carries the verbatim observation that earned it**. Unverifiable\ndimensions (auth-gated tooling) are marked *provisional*, never guessed.\nNo stars, no votes, no pay-to-rank — probes only, continuously re-run. Separate\nTrust Receipts publish dated security/access, data-integrity, citation,\nclaim-verification, response-benchmark, and reviewed field evidence without\ncollapsing it into a misleading trust score.\n\n**Agents:** connect to `https://mcpqueen.com/mcp` (streamable HTTP, no auth)\nand use `search_servers` to find working, graded servers for a task before\nyou commit to one. Machine surfaces: [`/api/grades.json`](https://mcpqueen.com/api/grades.json)\n· [`/api/changes.json`](https://mcpqueen.com/api/changes.json) ·\n[`/llms.txt`](https://mcpqueen.com/llms.txt). Setup guides:\n[`/integrations`](https://mcpqueen.com/integrations). Registry name:\n`com.mcpqueen/registry`.\n\n## Connect\n\nmcpqueen is a remote, no-auth, effectively **read-only** MCP server — safe to keep\nconnected as your discovery broker (only `submit_feedback` writes, and it just\nenqueues a quarantined field report). Ask your agent to `search_servers` for a task\n*before* it commits to an MCP.\n\n**Claude Code** (native HTTP):\n\n```\nclaude mcp add --transport http mcpqueen https://mcpqueen.com/mcp\n```\n\n**OpenClaw / Claude Desktop / any stdio client** — via the `mcp-remote` bridge; add\nto your `mcpServers` config (`~/.openclaw/openclaw.json`, `claude_desktop_config.json`, …):\n\n```json\n{\n \"mcpServers\": {\n \"mcpqueen\": {\n \"command\": \"npx\",\n \"args\": [\"-y\", \"mcp-remote\", \"https://mcpqueen.com/mcp\"]\n }\n }\n}\n```\n\n### OpenAI: ChatGPT and Codex\n\nMCP Queen can be registered directly as a private plugin/connector; no SDK,\nDocker image, or local command is required:\n\n1. In ChatGPT, open **Settings → Security and login** and enable\n **Developer mode**.\n2. Open [ChatGPT Plugins](https://chatgpt.com/plugins), select **+**, and choose\n the option to add an MCP server.\n3. Enter `https://mcpqueen.com/mcp` as a universal, no-auth remote MCP URL.\n4. Test with: “Find a well-maintained, no-auth MCP server for GitHub issue\n triage. Explain the evidence and any caveats.”\n\nFor an OpenAI Responses API demo, Node 18+ is enough:\n\n```bash\nexport OPENAI_API_KEY=\"your-api-key\"\nnpm run demo:openai\n```\n\nPass a custom prompt after `--`:\n\n```bash\nnpm run demo:openai -- \"Find an MCP server that can search FDA 510(k) records\"\n```\n\nThe demo allowlists only MCP Queen's read-only discovery and evidence tools.\n`submit_feedback` is intentionally excluded.\n\nFor public distribution in ChatGPT and Codex, create a **With MCP** submission\nin the [OpenAI plugin portal](https://platform.openai.com/plugins) and submit\nthe same universal endpoint. OpenAI's public review also requires verified\npublisher identity, public support/privacy/terms URLs, accurate tool safety\nannotations, starter prompts, and reviewer test cases.\nThe ready-to-paste listing copy and review cases are in\n[`docs/openai-submission.md`](docs/openai-submission.md). The timed recording\nplan, narration, captions, chapters, and visual assets are documented in the\n[`OpenAI demo production kit`](docs/openai-demo-production-kit.md).\n\n### Framework and agent examples\n\nRunnable examples are organized under [`examples/integrations`](examples/integrations):\n\n| Stack | Example | Needs a model key? |\n|---|---|---|\n| LangChain | [`MultiServerMCPClient`](examples/integrations/langchain) | No; calls a tool directly |\n| LlamaIndex | [`BasicMCPClient`](examples/integrations/llamaindex) | No; calls a tool directly |\n| Cloudflare Agents | [`Agent` + Workers AI](examples/integrations/cloudflare-agent) | No separate provider key |\n| Hugging Face | [`huggingface_hub.Agent`](examples/integrations/huggingface) | Yes, `HF_TOKEN` for inference |\n\nAll use the same public `https://mcpqueen.com/mcp` Streamable HTTP endpoint.\nThe agent examples exclude `submit_feedback` from automatic model access.\n\nBefore a release or directory submission, run the reusable artifact validator:\n\n```bash\nnpm run distribution:check\nnpm run distribution:check:live\n```\n\nIt checks the prepared package and live MCP surfaces. Publisher identity,\ndomain challenge tokens, demo recording, and final portal confirmations remain\nexplicit manual gates.\n\nThe measurable channel plan is in\n[`docs/distribution-strategy.md`](docs/distribution-strategy.md). Safe\nunattended maintenance and stop conditions are defined in\n[`docs/autonomous-operations.md`](docs/autonomous-operations.md) and\n[`AGENTS.md`](AGENTS.md).\n\n## Architecture (single Worker)\n\nSee the [system architecture and verification flow](docs/architecture.md) for\nthe ecosystem-level diagram and the Find → Verify → Connect decision loop.\n\n- `src/worker.ts` — everything: registry crawler, prober/grader, HTML pages,\n JSON API, and mcpqueen's **own MCP endpoint**.\n- `public/` — static landing (crown data-rain + Vex the fox) served via the\n assets binding; the Worker handles all non-asset routes.\n- D1 database `mcpqueen` (`schema.sql`): servers, probes, latest_grades,\n trust_observations, evidence_benchmark_runs, feedback (quarantined agent field\n reports), meta (sync cursor).\n- Cron `*/15 * * * *`: sync 4 registry pages + probe the 30 stalest remotes\n (~2,900 probes/day; full re-probe cycle ≈ 2.7 days over ~7.7K remotes).\n- Cron `17 7 * * *`: run one safe, read-only response audit against an eligible\n evidence/citation tool and publish dated results to its Trust Receipt.\n\n## Routes\n\n| Route | What |\n|---|---|\n| `/` | landing (static) |\n| `/registry` | leaderboard + methodology |\n| `/s/` | per-server grade with evidence + probe history |\n| `/api/grades.json` | grades as JSON (CORS open) |\n| `/mcp` | MCP server: capability discovery plus `get_trust_receipt` and `search_trust_evidence` |\n| `/integrations` | Setup matrix and runnable framework/agent examples |\n| `/field-reports` | Human-reviewed reports from agents that actually exercised a server |\n| `/api/trust/{name}.json` | Per-server operational, security, data-integrity, citation and claim evidence |\n| `/mcp-info` | for-agents page |\n| `/admin/*` | operator endpoints (key-gated) |\n\n\n## Grading rubric (deterministic, every point carries its observation)\n\nreachability 25 · protocol 15 · tooling 35 (tools/list, described %, typed %,\ndescription depth) · latency 10 · provenance 15 (metadata + namespace↔domain\nmatch). Auth-gated servers are scored on the verifiable subset and marked\n**provisional**. Agent feedback via `submit_feedback` is quarantined for human\nreview — never auto-published, never affects grades directly.\n\nTrust Receipts remain distinct from that grade. Safe response audits record\nusable-call rate, semantic upstream failures, returned PMID/DOI identifiers, and\nidentifier resolution against authoritative sources. Missing evidence is labeled\nunaudited rather than treated as a pass.\n\n## Deploy\n\n```\nnpm run deploy # wrangler deploy (any Cloudflare API token with Workers + D1 write)\nnpm run db:schema # apply schema.sql to remote D1\n```\n\nCustom domains mcpqueen.com + www are attached to the **Worker** (moved off the\nPages project 2026-07-12; mcpqueen.pages.dev still exists as a static preview of\n`public/` only — it has no /registry).\n", "bytes": 8026, "sha": "d2166de68d3edfd8cb2381128d583b856a8b4dea2e0cd8dc5c79baac010ec5a9", "repo_slug": "mcpqueen/mcpqueen", "fonte": "repo", "truncated": false, "api": "https://agentalog.com/api/listings/mcp_com_mcpqueen_registry_5575dd5e/readme" }