Medical MCP: drugs, guidelines and literature, no keys

Checking one drug can mean FDA and DailyMed for the US, TGA for Australia, EMA for Europe, then PubMed for the literature: five or six tabs before you have an answer. Medical MCP wires those public sources into one set of tools that runs locally and needs no API key to start. It had about 110 stars as of 2026-10-06, and v2.0 added circuit breakers, retries and rate limiting, so one failing source doesn’t sink the whole query.

What it does

• Drug information: search-drugs covers FDA, DailyMed, TGA, Health Canada and EMA with a country filter; search-drug-nomenclature normalizes names through RxNorm; search-drug-safety pulls FAERS adverse events, recalls and shortages.
• Literature: search-medical-literature searches PubMed and tags each hit with study type and evidence grade, from meta-analysis down to case report, with optional reranking by question.
• Guidelines and trials: search-clinical-guidelines for practice recommendations and search-clinical-trials for ClinicalTrials.gov.
• Pediatrics: separate tools for pediatric guidelines, pediatric journals and drugs with pediatric labeling.
• Health stats and self-checks: get-health-statistics reads the WHO Global Health Observatory, and health-check reports each source’s latency and breaker state.

Who it’s for

• Pharmacists, medical students and medical editors checking a drug’s label and safety warnings across countries.
• Content review teams that want an agent to sort the literature by evidence level first.

Setup

Needs Node.js 18+. This is the config behind the README’s one-click Cursor install, and it works in other clients too:

{
  "mcpServers": {
    "medical-mcp": {
      "command": "npx",
      "args": ["-y", "medical-mcp"]
    }
  }
}

An optional NCBI_API_KEY raises the PubMed limit from 3 to 10 requests per second.

Our take

It overlaps with BioMCP but serves different readers. BioMCP leans toward genomics and research; Medical MCP leans toward drug regulation and clinical literature, and querying several national regulators at once is its most practical trick. The README is candid that evidence grades are automatic labels drawn from titles and abstracts, not checked by a person, and that reranking is retrieval ordering, not diagnosis. Web sources like Google Scholar go through a third-party search only with MONID_API_KEY; otherwise it falls back to Semantic Scholar. It retrieves information and isn’t medical advice. MIT licensed.