The federation is the point: article_search queries five literature backends at once and deduplicates by identifier, and section-based fetching means a slow or broken source degrades one part of the answer instead of failing the call. Everything works keyless except OncoKB annotations and the DisGeNET path.
A biomedical MCP server that federates many upstream sources behind one set of tools — MyGene, MyVariant, MyChem, MyDisease, ClinVar, OpenTargets, OncoKB, DisGeNET, ClinicalTrials.gov, PubMed, EuropePMC and more. Searches fan out across backends and deduplicate, and a failed source degrades that section rather than the whole call.
- gene_search and gene_get — find a gene by symbol, name or keyword, and pull detail by section; smart mode resolves aliases such as HER2 to ERBB2
- gene_diseases, gene_drugs, gene_trials and gene_articles — the diseases associated with a gene, drugs targeting it, trials involving it, and the literature on it
- gene_enrich — pathway enrichment over a gene list
- variant_search, variant_get, variant_oncokb and variant_trials — search by rsid, HGVS or gene with clinical-significance and frequency filters, request extra sections on a hit, add OncoKB cancer annotations, and find trials for the variant
- drug_search, drug_get and drug_trials — drugs by name, mechanism or keyword, with regulatory, safety, target and indication sections
- disease_search, disease_get, disease_drugs and disease_trials — diseases by name or phenotype, with gene associations and phenotypes, plus their drugs and trials
- article_search and article_get — federated literature search across five backends with PMID, PMCID and DOI deduplication, and one article with open-access, annotation and citation sections
- trial_search and trial_get — clinical trials by condition, intervention or keyword, and one by NCT ID with eligibility, locations and outcomes
- pdb — search RCSB PDB structures, read entry metadata, and download mmCIF or PDB files
- discover and batch_get — resolve a free-text concept to entities, and fetch several entities in parallel
Nothing for the default path: npx biomcp starts a zero-config stdio server on Node.js 22.13 or newer. Two tools need credentials — variant_oncokb requires ONCOKB_TOKEN, and gene_diseases uses DISGENET_API_KEY when present, falling back to OpenTargets associations without it. Other optional keys raise rate limits.
One command — npx biomcp