NCBI

You can query PubMed’s vast literature through a dedicated MCP Server that translates natural language questions into precise PubMed searches. This server helps researchers in life sciences to perform literature reviews, discover relevant studies, and access complete article details efficiently, all through an MCP client interface.

How to use

You connect to the MCP server using an MCP client and run searches by describing your question in natural language. The server converts your request into PubMed queries, retrieves matching records, and returns abstracts, author lists, MeSH terms, DOIs, and publication details. You can refine results with advanced searches, look up related articles, and fetch full article details for papers of interest.

How to install

Prerequisites you need before starting are Python 3.8 or higher and Poetry (recommended) for dependency management.

1. Create and initialize the project.

mkdir ncbi-mcp-server && cd ncbi-mcp-server
poetry init

During initialization, add dependencies: mcp, httpx, and typing-extensions.

2. Create the project structure and save the server code.

mkdir -p src/ncbi_mcp_server
# Save server.py code as src/ncbi_mcp_server/server.py

3. Install dependencies.

poetry install

4. Test the server locally.

poetry run python src/ncbi_mcp_server/server.py

5. Configure the MCP client integration (Claude Desktop) by adding the MCP server entry to your Claude configuration. The example config below assumes your project lives at the path shown.

{
  "mcpServers": {
    "ncbi-literature": {
      "command": "poetry",
      "args": ["run", "python", "src/ncbi_mcp_server/server.py"],
      "cwd": "/FULL/PATH/TO/YOUR/ncbi-mcp-server"
    }
  }
}

Configuration and important notes

To run the server with the best performance, you can use an API key and email alignment for NCBI requests if you implement that in your environment. The server accepts local runtime execution via Poetry and a Python server script. You can deploy in development or production modes as needed.

Troubleshooting

If the server won’t start, ensure you are using Python 3.8 or newer and that all dependencies are installed with Poetry.

If searches return no results, verify your query syntax and consider using broader search terms or enabling dateRange controls in advanced searches.

If you encounter rate limit issues, obtain an NCBI API key and associate it with your requests to increase the limit per second.

Deployment quick reference

Local development and Docker deployment are supported. Use the provided shell commands to install dependencies and run the server, then connect via your MCP client.

The server is designed to be extended with additional data sources and search capabilities as your research needs grow.

Tools and endpoints you can use

You can access several core MCP functions to perform literature searches and retrieve article data.

Available tools

search_pubmed

Primary search tool for PubMed database; accepts query terms, supports field tags, and returns a list of PMIDs with basic metadata.

get_article_details

Fetch complete information for specific PubMed articles, including abstracts, authors, MeSH terms, DOI, and publication details.

search_mesh_terms

Find standardized MeSH terms to improve search precision and discover related concepts.

get_related_articles

Retrieve articles related to a given PubMed ID to aid literature reviews.

advanced_search

Perform complex queries by combining multiple terms, authors, journals, and publication types with logical operators.

get_analytics_summary

Provide an overview of usage, performance, and system health metrics for the MCP server.

get_detailed_metrics

Return detailed performance metrics for a specified time period, with hourly breakdowns and error rates.

reset_analytics

Permanently clear all collected analytics data.