What is BioMCP?
BioMCP, from the genomics software company GenomOncology, is a single command-line binary that is also an MCP server. One command grammar (search, get, pivot, enrich, batch) reaches 72 biomedical sources directly, including PubMed, PubTator3, Europe PMC, ClinVar, gnomAD, ClinicalTrials.gov, OncoKB, Reactome, UniProt and openFDA. An agent can move from a gene to its variants, trials, drugs and papers without rebuilding a query for each database, and a local study command analyses downloaded cBioPortal-style datasets.
It is a research aid, not medical advice. Its own privacy policy calls it a research and workflow aid and asks you to validate anything clinically relevant against primary sources, official labels and your institution's policies before patient care. The repository has about 670 stars and 119 forks, is MIT-licensed, written in Rust, and released v0.9.3 on 18 October 2026.
How to install BioMCP
uv tool install biomcp-cli && biomcp serveIs BioMCP worth it?
BioMCP's strength is that it knows how biomedical sources relate to each other. Asking about BRAF V600E returns a variant card with the ClinVar classification and gnomAD population frequencies in one read-only call, and helper commands pivot from that variant to its trials, papers or drugs. Article search fans out across PubTator3, Europe PMC and PubMed, removes duplicate PMIDs, PMCIDs and DOIs, and can add Semantic Scholar. JSON responses carry the likely next commands and the source behind each section, which gives an agent a provenance trail a person can check.
The question a clinician or a hospital's information governance lead should ask is what leaves the machine. According to the privacy policy, BioMCP has no telemetry, no analytics and no hosted service of its own: each query goes straight from your computer to the public providers needed to answer it, such as NCBI, ClinicalTrials.gov or openFDA, and those providers may log it. Whatever the tools return also goes to the model provider behind your MCP client. Responses are cached locally in a SQLite file for one day by default, with citation evidence kept for 30 days; --no-cache leaves no managed request state. The policy says not to send protected health information to third-party APIs without approval, and that applies here: a query naming a rare variant alongside a patient's age and town is identifying, whatever tool sends it.
The second thing to read is the licensing page. BioMCP itself is MIT, but the data it returns is governed by each source's terms. KEGG separates academic from non-academic use, COSMIC is reached only indirectly, and OncoKB needs a token. Version 0.9.3 added --disable-sources commercial-restricted, which switches off every source whose recorded terms restrict commercial use, a useful default for a company deployment.
Rate limits are per process: Semantic Scholar allows one request per second with a key and one every two seconds without. Teams running several agents are told to share one HTTP server, which has a Host check but no authentication or TLS, so it belongs behind an authenticated proxy. The release notes are long and specific, with 52 listed fixes in v0.9.3 alone, which reads as a project in heavy development rather than a finished one. For literature and variant research it is broader than any general web search server; for anything that reaches a patient it is a starting point that a qualified person must verify.
Where it earns its place
A postdoc in a cancer genomics lab
Gather the evidence on a variant before a lab meeting
Ask for the ClinVar significance, gnomAD frequency, linked papers and open trials for the variant, then follow the source links in the JSON output. It replaces an hour of tab-switching across five databases; the conclusions remain the postdoc's.
A clinical trials coordinator at a regional cancer centre
List recruiting trials for a tumour type and biomarker
Run a trial search on ClinicalTrials.gov filtered to recruiting, then pull eligibility and locations for the shortlist. Search only on disease and biomarker, never on patient details, and confirm eligibility with the trial sites.
A pharmacovigilance analyst at a mid-sized drug company
Summarise openFDA adverse-event signals for a marketed drug
Query FAERS counts and labels through the drug helpers with --disable-sources commercial-restricted set. Add a free openFDA key for higher rate limits, and check each source's terms before results go into a regulatory document.
Before you commit
- Setup
- Install the binary with the install script, uv tool install biomcp-cli (not the unrelated biomcp package), Homebrew or the Docker image, then register biomcp serve as a stdio server. A Claude Code plugin and a Codex command are documented; the README promises a first query in under 30 seconds.
- Integrations
- 72 direct sources across genes, variants, articles, trials, drugs, diseases, pathways, proteins, adverse events and pharmacogenomics. Optional keys for NCBI, Semantic Scholar, openFDA, NCI, OncoKB, DisGeNET and AlphaGenome. Embedded skill guides install into ~/.claude.
- Data and privacy
- Read-only. No telemetry, analytics or hosted control plane; queries go directly to the upstream providers, which may log them, and results go to your model provider. Local cache in biomcp.db, one day by default, citation evidence 30 days, user-only file permissions. The policy warns against sending protected health information to third-party APIs.
- Cost at scale
- Free under MIT. Costs come from your model tokens and any commercial data licences, for example KEGG for non-academic use or OncoKB for commercial work; prices for those are set by the providers and not published by BioMCP.
- Limits
- Rate limits are per process, so parallel agents should share one serve-http endpoint, which needs a proxy for authentication and TLS. Semantic Scholar references can be empty for paywalled papers. Batch requests take up to 10 IDs.
What’s good
- 72 sources behind one command grammar
- Read-only, with no telemetry or hosted service
- Section-level provenance in JSON output
- One switch turns off commercially restricted sources
What’s not
- Not a clinical decision tool; output needs expert checking
- Queries and results reach upstream providers and your model provider
- Shared HTTP mode has no built-in authentication
- Several valuable sources need separate keys or licences
Best for
Researchers, bioinformaticians and clinical informatics teams who want an agent to gather literature, variant and trial evidence with sources attached.
BioMCP alternatives
BioMCP: common questions
- Is BioMCP worth it?
- Gives an agent the public biomedical literature, variant and trial databases with every answer sourced, and its output is research leads, never clinical advice. We rate it 4.2 out of 5.
- How much does BioMCP cost?
- Free and open source (MIT). Most sources need no key; OncoKB, DisGeNET, NCI and AlphaGenome features need your own credentials, and some sources restrict commercial use.
- What are the downsides of BioMCP?
- Not a clinical decision tool; output needs expert checking; queries and results reach upstream providers and your model provider; shared HTTP mode has no built-in authentication; several valuable sources need separate keys or licences.
- Who is BioMCP best for?
- Researchers, bioinformaticians and clinical informatics teams who want an agent to gather literature, variant and trial evidence with sources attached.
- How much does BioMCP cost for a team or in production?
- Free under MIT. Costs come from your model tokens and any commercial data licences, for example KEGG for non-academic use or OncoKB for commercial work; prices for those are set by the providers and not published by BioMCP.
- Is BioMCP safe for company data?
- Read-only. No telemetry, analytics or hosted control plane; queries go directly to the upstream providers, which may log them, and results go to your model provider. Local cache in biomcp.db, one day by default, citation evidence 30 days, user-only file permissions. The policy warns against sending protected health information to third-party APIs.
- What are the best BioMCP alternatives?
- The closest alternatives we have reviewed are Zotero MCP and Exa MCP Server.
Answers checked Nov 7, 2026.
Make BioMCP? Show this rating on your site
