pubmed-search

PubMed Search MCP is an intelligent middleware designed to help AI agents perform literature search and analysis across multiple scholarly sources. Built around Domain-Driven Design, it offers a suite of tools and workflows to translate agent intents into precise search queries, route requests to PubMed, Europe PMC, CORE, and other data sources, and provide structured outputs suitable for downstream reasoning. The server ships with a broad set of capabilities, including PICO-driven query construction, auto MeSH/ ICD mappings, disambiguation, and multi-source full-text access with export options. It is designed to be agent-first, returning decision-friendly results and actionable insights rather than raw human-readable pages. To run it, configure the MCP client to start the server over stdio using uvx, with the required environment variable NCBI_EMAIL set as part of policy compliance.

Once running, you can leverage its unified search pathways, research timeline features, citation network analysis, and one-click export options to retrieve RIS, BibTeX, CSV, or MEDLINE formats, simplifying integration with agent workflows or external tooling.

Prerequisites:\n- Python 3.10+ (recommended)\n- uv (recommended) — install via the provided script in the README: curl -LsSf https://astral.sh/uv/install.sh | sh (macOS/Linux) or the PowerShell command for Windows.\n- An NCBI_EMAIL address (required by NCBI API policy). Optional: NCBI_API_KEY for higher rate limits.\n\nInstallation & Run Options:\n1) Zero-install with uvx (recommended for trying out):\n``` uvx pubmed-search-mcp

2) Add as a project dependency:\n```
uv add pubmed-search-mcp

3. pip install (if you prefer Python packaging):\n``` pip install pubmed-search-mcp

\nEnvironment note: If you are using stdio transport, ensure NCBI_EMAIL is set in your environment before starting the server. For example:\n```
NCBI_EMAIL=your@email.com uvx pubmed-search-mcp

Tips and troubleshooting:\n- NCBI_EMAIL is required by the API policy and must be set for all requests. Optionally set NCBI_API_KEY to increase rate limits (10 req/s vs 3 req/s).\n- The server supports multiple MCP clients via stdio; configure your client to point to the uvx command with the pubmed-search-mcp name.\n- If you encounter connection or transport issues, verify that the environment variables are exported in the shell that launches uvx.\n- For advanced deployments, consider using the OpenCLI/VS Code/Claude/Zed integration snippets provided in the README to add the server to your preferred client.\n- Detailed integration guides covering environment variables, Copilot Studio setup, Docker deployment, proxy configuration, and troubleshooting are available in the project docs.\n- When using multiple data sources, expect a unified search experience with auto-disambiguation, PICO-driven query generation, and parallelized querying with deduplication.