| name | biomcp-server |
| description | MCP bio bridge |
| keywords | ["MCP","PubMed","ClinicalTrials","server","uv"] |
| measurable_outcome | Stand up a working BioMCP endpoint (pip or uv) and return ≥1 PubMed + ≥1 ClinicalTrials.gov response to the client within 10 minutes. |
| license | MIT |
| metadata | {"author":"BioMCP Team","version":"1.0.0"} |
| compatibility | [{"system":"MCP-compliant clients"}] |
| allowed-tools | ["web_fetch"] |
BioMCP Server
Deploy and operate the BioMCP server so MCP-compatible clients (Claude Desktop, LobeChat, etc.) can query biomedical databases via a single standardized interface.
When to Use
- Unified literature search (PubMed/PMC) inside MCP clients.
- Entity normalization via PubTator3 or genomic variant lookups.
- ClinicalTrials.gov queries without bespoke API wrappers.
Core Capabilities
- PubMed/PMC search: Execute complex literature queries.
- PubTator3 annotations: Map text to genes, diseases, chemicals, species.
- ClinicalTrials.gov: Retrieve trial metadata/protocols.
- Genomic variant lookups: Fetch variant/gene summaries from connected sources.
Deployment Workflow
- Install deps:
cd repo && uv sync (preferred) or pip install ..
- Run server:
python -m biomcp.server or make run; Docker Compose provided.
- Configure client: Add command/args snippet from
README.md into MCP client config (Claude Desktop, BioKernel, etc.).
- Test tools: Invoke PubMed + ClinicalTrials + variant endpoints to ensure connectivity.
- Monitor: Capture logs, rate-limit statuses, and data-source versions for audit.
Guardrails
- Keep API keys/env secrets outside the repo.
- Respect upstream rate limits to avoid throttling or bans.
- Document which data sources are enabled per deployment and update when they change.
References
- Source repo + configuration examples in
README.md, repo/docker-compose.yml, and repo/Makefile.