PubTator MCP Server

🔍 A biomedical literature annotation and relationship mining server based on PubTator3, providing convenient access through the MCP interface.

PubTator MCP Server provides AI assistants with access to the PubTator3 biomedical literature annotation system through the Model Context Protocol (MCP). It allows AI models to programmatically search scientific literature, obtain annotation information, and analyze entity relationships.

🤝 Contribute • 📝 Report Issues

✨ Core Features

🔎 Literature Annotation Export: Support exporting PubTator annotation results in multiple formats ✅
🚀 Entity ID Lookup: Query standard identifiers for biological concepts through free text ✅
📊 Relationship Mining: Discover biomedical relationships between entities ✅
📄 Literature Search: Support literature retrieval by keywords and entity IDs ✅
🧠 Batch Processing: Support batch export of annotation information from search results ✅

🚀 Quick Start

Requirements

Python 3.10+
FastMCP library

Installation

Via Smithery

Use Smithery to automatically install PubTator Server:

Claude

npx -y @smithery/cli@latest install @JackKuo666/pubtator-mcp-server --client claude --config "{}"

Cursor

Paste in Settings → Cursor Settings → MCP → Add new server:

Mac/Linux

npx -y @smithery/cli@latest run @JackKuo666/pubtator-mcp-server --client cursor --config "{}"

Windsurf

npx -y @smithery/cli@latest install @JackKuo666/pubtator-mcp-server --client windsurf --config "{}"

CLine

npx -y @smithery/cli@latest install @JackKuo666/pubtator-mcp-server --client cline --config "{}"

Manual Installation

Clone the repository:

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

Install dependencies:
```
pip install -r requirements.txt
```

📊 Usage

Running the Server Directly

Start the MCP server:

python pubtator_server.py

The server now supports both stdio and TCP transports. By default, it uses TCP transport. You can configure the following environment variables:

MCP_TRANSPORT: Set to "tcp" for TCP transport or "stdio" for stdio transport (default is "tcp")
MCP_HOST: The host to bind to (default is "0.0.0.0")
MCP_PORT: The port to listen on (default is 8080)

Example of starting the server with custom settings:

MCP_TRANSPORT=tcp MCP_HOST=127.0.0.1 MCP_PORT=8888 python pubtator_server.py

The server implements lazy initialization and proper error handling. It will gracefully handle shutdown signals (SIGINT and SIGTERM) and log any errors that occur during startup or operation.

Using Docker

We provide a Dockerfile for easy deployment. To use the Docker container:

Build the Docker image:
```
docker build -t pubtator-mcp-server .
```

Run the Docker container:

docker run -p 8080:8080 pubtator-mcp-server

This will start the PubTator MCP server inside a Docker container, exposing it on port 8080.

Troubleshooting

If you encounter any issues starting the server:

Check the console output for error messages.
Ensure all required dependencies are installed (see Requirements section).
Verify that the environment variables are set correctly.
If the server fails to start, try running it with increased verbosity:

python -v pubtator_server.py

This will provide more detailed logging information to help identify the source of any issues.

When using Docker, you can check the logs with:

docker logs <container_id>

Configuration

Claude Desktop Configuration

Add to claude_desktop_config.json:

(Mac OS)

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

(Windows)

{
  "mcpServers": {
    "pubtator": {
      "command": "C:\\Users\\YOUR\\PATH\\miniconda3\\envs\\mcp_server\\python.exe",
      "args": [
        "D:\\code\\YOUR\\PATH\\PubTator-MCP-Server\\pubtator_server.py"
      ],
      "env": {},
      "disabled": false,
      "autoApprove": []
    }
  }
}

CLine Configuration

{
  "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": []
    }
  }
}

To use TCP transport, modify the configuration as follows:

{
  "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": "tcp",
        "MCP_HOST": "127.0.0.1",
        "MCP_PORT": "8888"
      },
      "disabled": false,
      "autoApprove": []
    }
  }
}

🛠 API Features

PubTator MCP Server provides the following core features:

1. Export Publications (export_publications)

Export PubTator annotation results for specified PMID literature:

result = await export_publications(
    ids=["32133824", "34170578"],
    id_type="pmid",
    format="biocjson",  # Supported: pubtator, biocxml, biocjson
    full_text=False     # Whether to include full text
)

2. Entity ID Lookup (find_entity_id)

Query standard identifiers for biological concepts through free text:

result = await find_entity_id(
    query="COVID-19",
    concept="disease",  # Optional: gene, disease, chemical, species, mutation
    limit=5             # Optional: limit number of results
)

3. Relationship Query (find_related_entities)

Find other entities related to a specified entity:

result = await find_related_entities(
    entity_id="@DISEASE_COVID_19",
    relation_type="treat",    # Optional: treat, cause, interact, etc.
    target_entity_type="chemical",  # Optional: gene, disease, chemical
    max_results=5       # Optional: limit number of results
)

4. Literature Search (search_pubtator)

Search the PubTator database:

results = await search_pubtator(
    query="COVID-19",
    max_pages=1     # Optional: maximum number of pages to retrieve
)

5. Batch Export (batch_export_from_search)

Search and batch export literature annotations:

results = await batch_export_from_search(
    query="COVID-19",
    format="biocjson",
    max_pages=1,
    full_text=False,
    batch_size=5
)

Note: The actual function calls may vary depending on your implementation. These examples are based on our recent tests and may need to be adjusted to match your exact API.

⚠️ Usage Limitations

API request rate limit: maximum 3 requests per second
When batch exporting, use a reasonable batch_size to avoid request timeout
For relationship queries, entity IDs must start with "@", e.g., "@DISEASE_COVID-19"

📄 License

This project is licensed under the MIT License.

⚠️ Disclaimer

This tool is for research purposes only. Please comply with PubTator's terms of service and use this tool responsibly.