MCP server

EvidenceLens implements the Anthropic Model Context Protocol so any MCP-aware client — Claude, ChatGPT, IDE plugins, custom agents — can call its evidence tools directly. The server is public and needs no API key.

Endpoint

Remote (Streamable HTTP): https://evidencelens.mykpoplists.com/mcp
This is the modern MCP transport (a single endpoint), which both Claude.ai custom connectors and ChatGPT require. A legacy HTTP+SSE endpoint also exists at /sse for older clients, but new integrations should use /mcp.
Local (stdio): build from mcp-server/ and run node dist/server.js (set GATEWAY_URL=https://evidencelens.mykpoplists.com).

Test it free with the MCP Inspector

No account needed. With Node.js installed:

Run npx @modelcontextprotocol/inspector (it opens a browser tab; use the printed localhost:6274 link with its token).
Set Transport Type to Streamable HTTP and paste the endpoint above as the URL, then Connect.
Tools → List Tools, pick search_evidence, run it with { "query": "aspirin cardiovascular", "top_k": 5 }.

Add as a connector

Claude.ai (free tier allows one): Settings → Connectors → Add custom connector → paste https://evidencelens.mykpoplists.com/mcp.
ChatGPT / Claude Desktop / Cursor / Cline: add a remote MCP server with the same URL.

Verify with curl

curl -X POST https://evidencelens.mykpoplists.com/mcp \
  -H 'content-type: application/json' \
  -H 'accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize",
       "params":{"protocolVersion":"2025-06-18","capabilities":{},
       "clientInfo":{"name":"curl","version":"0"}}}'

Tools

search_evidence — hybrid biomedical search with COI metadata + facet filters.
get_paper — fetch one document by canonical id (with citation neighborhood).
get_trial — single ClinicalTrials.gov / ICTRP record.
get_trials_by_condition — trials by condition, location, status, phase.
get_recent_recalls — recent FDA / EMA recall events.
get_author_payments — CMS Open Payments records for an author.
get_citation_neighborhood — walk the citation graph from one document.
evaluate_evidence_quality — per-document evidence-quality scorecard.

Rate limits

30 tool calls per minute per client. Returns an MCP isError response with a retry hint when exceeded.

Resources

Documents are also exposed as MCP resources under the URI template evidencelens://document/{id}, so clients that prefer the resource interface over tool calls can read them directly.