aureten/targetval-mcp-server

TargetVal MCP Server (remote)

This is a minimal remote MCP server that wraps your TargetVal Gateway and exposes the two tools required by ChatGPT connectors and Deep Research:

• search(query: str) → returns a JSON string {"results":[{"id","title","url"}...]}
• fetch(id: str) → returns a JSON string {"id","title","text","url","metadata"}

It maps free-form queries like PCSK9 in hypercholesterolemia to a handful of TargetVal endpoints (aggregate, PPI, pathways, known drugs, clinical endpoints), and lets ChatGPT retrieve the full JSON via fetch.

Deploy on Render

1. Create a new Web Service in Render, connect this repo/folder, or upload the ZIP.
2. Build command: pip install -r requirements.txt
3. Start command: (from Procfile) web: python mcp_server.py
4. Set env vars (optional):
   • TARGETVAL_BASE=https://targetval-gateway.onrender.com (default)
   • TARGETVAL_LIMIT=25
   • MCP_TRANSPORT=sse (or http)
   • MCP_PATH=/sse
5. Visit https://<your-app>.onrender.com/sse (or /mcp) to verify the endpoint is reachable.

Connect in ChatGPT (Developer mode)

1. Open Settings → Connectors → Advanced → Developer mode and enable it.
2. Add a new Remote MCP server:
   • Server URL: https://<your-app>.onrender.com/sse (or /mcp if you used HTTP transport)
   • Allowed tools: search, fetch
   • Require approval: never
3. In a chat, choose Use Connectors and select your server.
4. Ask: PCSK9 in hypercholesterolemia

API (Deep Research / Responses API)

Use the Responses API with an MCP tool config like:

curl https://api.openai.com/v1/responses   -H "Content-Type: application/json"   -H "Authorization: Bearer $OPENAI_API_KEY"   -d '{
    "model": "o4-mini-deep-research",
    "input": [{"role":"user","content":[{"type":"input_text","text":"PCSK9 in hypercholesterolemia — summarize evidence"}]}],
    "tools": [{"type":"mcp","server_label":"targetval","server_url":"https://<your-app>.onrender.com/sse","allowed_tools":["search","fetch"],"require_approval":"never"}]
  }'

How IDs work

search() returns opaque IDs that encode the HTTP method and request target:

• GET|https://.../mech/ppi?symbol=PCSK9&cutoff=0.9&limit=25
• POST|https://.../aggregate|{"symbol":"PCSK9","condition":"hypercholesterolemia","modules":["mech_ppi","mech_pathways","tract_drugs","clin_endpoints"],"limit":25}

fetch(id) parses these and performs the HTTP call to your TargetVal Gateway, then wraps the JSON payload into the expected MCP shape.

Notes

• Keep limits modest to avoid oversized tool responses.
• You can extend _results_for() to add more modules or tweak defaults.
• If /aggregate is heavy in your deployment, stick to per-module GETs.