PubMed MCP

A lightweight stdio MCP server for PubMed search via the NCBI E-utilities API.

This open-source copy removes machine-specific paths and does not include any credentials.

Features

• search_pubmed
• Returns up to 50 records per request
• Output includes pmid, title, and abstract
• No dependency on the Python mcp package

Requirements

• Python 3.10+
• requests
• An NCBI API key

Installation

python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt

Authentication

By default the server reads credentials from ./auth.json.

You can also override the path with PUBMED_AUTH_FILE.

1. Copy auth.example.json to auth.json
2. Fill in your own credentials
3. Do not commit auth.json

Example:

{
  "api_key": "your_ncbi_api_key",
  "email": "your_email@example.com"
}

Run

python server.py

Codex MCP Configuration

[mcp_servers.pubmed-search]
command = "python"
args = ["/path/to/pubmed-mcp/server.py"]

[mcp_servers.pubmed-search.env]
PUBMED_AUTH_FILE = "/path/to/pubmed-mcp/auth.json"

Notes

• The server only reads credentials locally and never returns them in tool output.
• limit must be between 1 and 50.
• The default upstream endpoint is https://eutils.ncbi.nlm.nih.gov/entrez/eutils.

Files

• server.py: MCP server implementation
• auth.example.json: local credential template
• README_zh_cn.md: Chinese documentation
• LICENSE: MIT license

License

MIT. See LICENSE.