Medical MCP

What JamesANZ/medical-mcp MCP server does

JamesANZ/medical-mcp MCP server exposes medical information sources to MCP-compatible AI clients. It supports searches for FDA drug records, RxNorm terminology, WHO health indicators, PubMed articles, Google Scholar results, clinical guidelines, and pediatric resources. The available data is intended for tasks such as literature discovery, drug-reference lookups, health-statistics research, and finding recommendations from medical organizations.

The project runs locally and does not require API keys. Its README describes localhost-only operation, no query storage, and no tracking or cloud service operated by the project. External source requests are made as needed, while returned API data can be retained in an in-memory cache.

How it works

The server is built with Node.js and TypeScript and uses the MCP SDK to expose named tools. Tools accept search or lookup arguments, such as a drug query, NDC code, PMID, health indicator, country, or literature topic. Results are retrieved from the relevant source and returned to the connected AI client.

Available source coverage includes FDA drug data, WHO Global Health Observatory statistics, PubMed citations, RxNorm nomenclature, Google Scholar, and multi-database searches spanning PubMed, Scholar, Cochrane, and ClinicalTrials.gov. Pediatric functions cover AAP and Bright Futures guidance, pediatric journals, child-health indicators, and drugs with pediatric labeling or dosing information.

Responses use an in-memory cache with source-specific expiration periods. The cache also removes expired entries periodically and applies least-recently-used eviction after reaching its configured maximum size. The get-cache-stats tool reports cache-related information such as hit rate, memory usage, and entry count.

Setup and configuration

JamesANZ/medical-mcp MCP server requires Node.js 18 or newer and npm. The README provides a Cursor one-click installation link and a manual global installation using the medical-mcp npm package. When building from source, clone the repository, run npm install, build with npm run build, and start it with npm start.

Claude Desktop can launch the built JavaScript entry point through its claude_desktop_config.json file. Cursor and Claude Desktop are the clients explicitly documented by the project. The repository also lists macOS, Windows, and Linux as supported platforms.

Optional cache settings include CACHE_ENABLED, CACHE_MAX_SIZE, source-specific TTL variables such as CACHE_TTL_FDA, CACHE_TTL_PUBMED, CACHE_TTL_WHO, and CACHE_TTL_RXNORM, plus CACHE_CLEANUP_INTERVAL. The documented defaults enable caching, allow up to 1,000 entries, and use different lifetimes for each source.

Tools and capabilities

The server provides tools for:

• Searching FDA drugs, retrieving drug details by NDC, and looking up RxNorm names.
• Reading WHO health statistics such as life expectancy, mortality, and disease prevalence.
• Searching PubMed, Google Scholar, medical journals, and multiple medical databases.
• Finding clinical guidelines, AAP guidance, Bright Futures material, and pediatric literature.
• Retrieving pediatric drug information and WHO child-health indicators.
• Inspecting cache statistics.

Google Scholar access uses web scraping with rate limiting; the README states that other sources use official APIs.

Limitations and notes

The project does not replace medical advice, diagnosis, or treatment. Medical decisions should be reviewed with qualified healthcare professionals. Source freshness varies: the README describes daily PubMed and pediatric-journal updates, annual WHO statistics, weekly RxNorm updates, and periodic AAP updates, while some sources are described as real-time.

Google Scholar may be subject to scraping and rate-limit behavior. Local execution protects the server's process and query handling, but the server still contacts external medical sources to retrieve current results. No credentials are required according to the project documentation.