Overview
What is PubTator MCP Server?
It is a biomedical literature annotation and relationship mining server based on PubTator3, providing AI assistants access to PubTator3 through the Model Context Protocol (MCP). It allows AI models to programmatically search scientific literature, obtain annotation information, and analyze entity relationships.
How to use PubTator MCP Server?
Install via Smithery (npx command) or manually clone the repo and run pip install -r requirements.txt. Start the server with python pubtator_server.py. Configure transports via environment variables MCP_TRANSPORT, MCP_HOST, MCP_PORT. Use the MCP client (e.g., Claude Desktop, Cursor) to connect and call tools like export_publications, find_entity_id, find_related_entities, search_pubtator, and batch_export_from_search.
Key features of PubTator MCP Server
- Export PubTator annotations in multiple formats (pubtator, biocxml, biocjson)
- Look up standard identifiers for biological concepts via free text
- Discover biomedical relationships (treat, cause, interact, etc.)
- Search literature by keywords and entity IDs
- Batch export annotations from search results
Use cases of PubTator MCP Server
- Retrieve annotated publications for given PMIDs for analysis
- Find the official ID of a gene, disease, or chemical from a natural language query
- Identify entities related to a known biomedical concept
- Search literature on a topic and export all annotations in batch
- Integrate literature mining directly into AI assistant workflows
FAQ from PubTator MCP Server
What are the rate limits?
API requests are limited to a maximum of 3 requests per second.
What data formats are supported for annotation export?
Supported formats are pubtator, biocxml, and biocjson.
What transport options are available?
The server supports both stdio and TCP transport, configurable via the MCP_TRANSPORT environment variable (default is TCP).
Are there any special requirements for entity IDs in relationship queries?
Entity IDs must start with "@", for example @DISEASE_COVID-19.
What are the runtime dependencies?
Python 3.10+ and the FastMCP library are required.
