PubTator

You can access PubTator’s biomedical literature annotation and relationship mining capabilities through an MCP server. This lets AI models programmatically search literature, retrieve annotations, and explore entity relationships via a simple, client-driven interface.

How to use

Connect with the PubTator MCP Server using an MCP client you already run in your environment. You can perform common tasks such as exporting PubTator annotations for specific papers, looking up standard entity identifiers from free text, querying relationships between entities, searching PubTator for literature, and conducting batch exports from search results.

Practical usage patterns include exporting annotation results for a list of PMIDs, resolving terms like a disease or chemical name to standard IDs, discovering related entities for a given concept, and performing literature searches by keywords or IDs. When you work with batch exports, balance the batch size with your network and server capacity to avoid timeouts.

How to install

Prerequisites you need on your system before starting are Python 3.10 or newer and the FastMCP library. You will also clone the server source and install dependencies as part of the setup.

Step 1. Clone the server repository

git clone https://github.com/JackKuo666/PubTator-MCP-Server.git
cd PubTator-MCP-Server

Step 2. Install Python dependencies

pip install -r requirements.txt

Step 3. Start the server directly or via a container (see the Configuration section for client-specific start commands and environment settings)

Configuration

You can run the PubTator MCP Server in a local stdio mode or expose it over TCP. The following configuration examples show how to start the server via standard input/output (stdio) using explicit commands from a couple of common setup methods.

{
  "mcpServers": {
    "pubtator": {
      "command": "python",
      "args": ["-m", "pubtator-mcp-server"]
      }
  }
}

{
  "mcpServers": {
    "pubtator": {
      "command": "bash",
      "args": [
        "-c",
        "source /home/YOUR/PATH/mcp-server-pubtator/.venv/bin/activate && python /home/YOUR/PATH/pubtator_server.py"
      ],
      "env": {
        "MCP_TRANSPORT": "stdio"
      },
      "disabled": false,
      "autoApprove": []
    }
  }
}

Usage notes

The server supports both stdio and TCP transports. By default, TCP is used, but you can switch to stdio by adjusting the MCP_TRANSPORT setting in the environment. If you choose TCP, you should configure MCP_HOST and MCP_PORT to bind the desired address and port.

Available tools

export_publications

Export PubTator annotation results for specified literature identifiers in common formats such as pubtator, biocxml, or biocjson.

find_entity_id

Query standard identifiers for biological concepts from free text, with optional concept filtering and result limit.

find_related_entities

Discover entities related to a given entity along with the relationship type and optional target type.

search_pubtator

Perform literature searches within the PubTator database using keywords or IDs.

batch_export_from_search

Search and batch export literature annotations with a specified format and batch configuration.