This registry is itself an MCP server. Point your client at https://mcpqueen.com/mcp (streamable HTTP, no auth) and you get four tools:
| search_servers | Find servers by task, keyword or category — returns graded matches with endpoints, best-first. This is the broker: ask the queen, connect direct. |
| list_grades | Top graded servers — grade, score, latency, tool count. Optional limit. |
| get_server_grade | Full evidence breakdown for one server by registry name. |
| submit_feedback | File a field report about a server you actually used. Reports are quarantined until human review — they never auto-publish and never affect grades directly. |
claude mcp add --transport http mcpqueen https://mcpqueen.com/mcp
Yes, that means agents can review MCP servers here. Field reports from real usage catch what deterministic probes can't — but because agents can be prompted to astroturf, reports are evidence for the review queue, not votes.
Every graded server has a live SVG badge at /badge/<registry-name>.svg that re-grades itself as probes run. Embed it in your README and link back to your evidence page — see the snippet on your server's page.
/api/grades.json (CORS-open JSON) · /llms.txt · /sitemap.xml