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

1. PubMed/PMC search: Execute complex literature queries.
2. PubTator3 annotations: Map text to genes, diseases, chemicals, species.
3. ClinicalTrials.gov: Retrieve trial metadata/protocols.
4. Genomic variant lookups: Fetch variant/gene summaries from connected sources.

Deployment Workflow

1. Install deps: cd repo && uv sync (preferred) or pip install ..
2. Run server: python -m biomcp.server or make run; Docker Compose provided.
3. Configure client: Add command/args snippet from README.md into MCP client config (Claude Desktop, BioKernel, etc.).
4. Test tools: Invoke PubMed + ClinicalTrials + variant endpoints to ensure connectivity.
5. 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.