BioMCP: one grammar across 70 biomedical sources

Looking up a single variant usually means hopping between ClinVar, gnomAD, OncoKB, PubMed and ClinicalTrials.gov, each with its own identifiers and search habits. BioMCP puts those sources behind one command grammar, and the same binary works in a terminal or as an MCP server for Claude Code, Codex or Claude Desktop. It’s maintained by GenomOncology and had about 650 stars as of 2026-10-06.

What it does

• One grammar: a handful of verbs (search, get, batch, enrich) span genes, variants, articles, trials, drugs, diseases, pathways, proteins, adverse events and pharmacogenomics, as in biomcp get variant "BRAF V600E" clinvar.
• Pivots: jump from a gene straight to its variants, drugs or diseases without rebuilding filters; search all gives counts across entity types first.
• Literature trails: search article queries PubTator3 and Europe PMC and dedupes the hits; article citations and article references expand one paper into its citation network.
• Local study analysis: study commands run cohort, survival and co-occurrence analysis on downloaded cBioPortal-style datasets, with terminal, SVG or PNG charts.
• Playbooks: biomcp skill list shows bundled investigation workflows you can install into an agent’s directory.

Who it’s for

• Cancer genomics and precision medicine researchers who need the evidence on a variant fast.
• Bioinformatics engineers who’d rather have an agent make the API calls than maintain a pile of scripts.
• Trial coordinators filtering studies by condition and recruiting status.

Setup

Install the biomcp binary, then register it. The PyPI package is biomcp-cli; a package named biomcp is unrelated:

uv tool install biomcp-cli

{
  "mcpServers": {
    "biomcp": {
      "command": "biomcp",
      "args": ["serve"]
    }
  }
}

Homebrew, a Docker image and a remote biomcp serve-http mode are also available.

Our take

Plenty of biomedical MCP servers wrap one or two APIs. BioMCP covers far more ground and is better engineered: a single Rust binary, a source table that says which provider backs each entity, and docs that steer agents away from inventing get commands that don’t exist. Two caveats. The data comes from many upstream providers, and some restrict commercial or clinical use, so check their terms before relying on it. And it’s a research lookup tool, not a substitute for a clinician’s diagnosis or treatment advice. Some sources, such as Semantic Scholar, work better with an API key. MIT licensed.