PubMed MCP

PubMed MCP exposes the PubMed E-Utilities search workflow over HTTP so it can be consumed by clients that implement the Model Context Protocol (MCP) or any other REST-capable integration. It provides a single endpoint for querying PubMed and returns curated article metadata in a consistent JSON shape.

Features

• 🔎 Search PubMed with a single HTTP request.

• 📄 Retrieves article summaries (title, journal, publication date, authors, DOI, and canonical URL).

• 🛠️ Configurable via environment variables, including an optional NCBI API key for higher rate limits.

• 🧪 Tested with mocked responses to ensure resilient parsing of PubMed responses.

Related MCP server: PubMed MCP Server

Getting started

1. Install dependencies

Create and activate a Python 3.10+ environment, then install the dependencies:

pip install -e .
# For development and testing extras
pip install -e .[dev]

2. (Optional) Configure an API key

Set PUBMED_API_KEY or NCBI_API_KEY with your NCBI E-Utilities API key to unlock higher throughput.

export PUBMED_API_KEY="your-api-key"

3. Launch the HTTP server

Run the FastAPI app with Uvicorn:

uvicorn pubmed_mcp.server:app --host 0.0.0.0 --port 8000
# or
python -m pubmed_mcp.main

You can now query the service:

curl "http://localhost:8000/articles/search?q=heart+disease&limit=5"

4. Run the test suite

pytest

API

• GET /health – Returns { "status": "ok" } for readiness checks.

• GET /articles/search – Required query parameter q; optional limit (default 10, max 50). Returns a JSON payload containing the original query, the applied limit, and a list of article summaries.

Development notes

• The PubMedClient handles eSearch and eSummary requests and merges them into a list of ArticleSummary dataclasses.

• The FastAPI app caches a single client instance and gracefully closes it during shutdown.

• Tests use respx to simulate PubMed responses, keeping the suite fast and deterministic.