BioContextAI Knowledgebase MCP

BioContextAI - Knowledgebase MCP

A Model Context Protocol (MCP) server for biomedical research that provides a standardized connection layer between artificial intelligence systems and biomedical resources. Documentation and usage guides are available at: https://biocontext.ai

Overview

BioContextAI Knowledgebase MCP is an MCP server implementation for common biomedical resources, enabling agentic large language models (LLMs) to retrieve verified information and perform domain-specific tasks. Unlike previous approaches that required custom integration for each resource, BioContextAI KB MCP provides a unified access layer through the Model Context Protocol that enables interoperability between AI systems and domain-specific data sources.

The Knowledgebase MCP is available both as:

• An open-source software package for local hosting (see Installation) - Choose this option for Claude Desktop, IDEs or your own agentic systems
• A remote server for setup-free integration at https://biocontext-kb.fastmcp.app/mcp (for testing purposes only, subject to fair use)

[!WARNING] If possible, we encourage you to run BioContextAI Knowledgebase MCP locally to avoid rate limits and ensure the service's availability for applications that rely on remote hosting.

The Knowledgebase MCP is part of the wider BioContextAI project. The BioContextAI Registry catalogues community servers that expose biomedical databases and analysis tools, providing the community with a resource for tool discovery and distribution. The registry index can be found at: https://biocontext.ai/registry.

Implemented Tools

BioContextAI Knowledgebase MCP exposes a number of external biomedical APIs. You can think of BioContextAI as a browser for your LLM that allows it to find relevant information across these knowledge bases. Please make sure to adhere to the usage limits (e.g., rate limits) of the respective services when using BioContextAI Knowledgebase MCP. If you use data from these services in your research, please make sure to cite both BioContextAI as well as the respective data source/tool.

[!WARNING] The data accessed through these APIs is not covered by the BioContextAI Knowledgebase MCP license. You are responsible for ensuring that your use of the data aligns with permitted practices.

Tools

• Antibody Registry - Gene id conversion
• bioRxiv/medRxiv - Recent preprint search and metadata access
• Ensembl - Gene id conversion
• EuropePMC - Literature search and full-text access
• Google Scholar - Academic publication and author search (only available for local use due to rate limiting)
• InterPro - Protein families, domains, and functional sites classification
• KEGG - Pathways, gene and drug-drug interaction database (only available for local use due to licensing restrictions)
• OpenTargets - Target-disease associations
• PanglaoDB - Single-cell RNA-sequencing cell type markers
• PRIDE - Proteomics data repository for mass spectrometry data
• Protein Atlas - Protein expression data
• Reactome - Pathways database
• STRING - Protein-protein interaction networks
• AlphaFold DB - Tertiary protein structure predictions

OpenAPI MCP Servers

FastMCP allows for easy conversion of REST endpoints following the OpenAPI specification into MCP servers. We have added code to automatically create such servers based on schemas provided through a configuration file, so that users deploying their own version of BioContextAI can easily extend the list of available tools. The configuration file is located at src/biocontext_kb/openapi/config.yaml. By default, no OpenAPI servers are included, but you can edit the configuration file to add services.

Installation

1. Local setup with Claude Desktop:

Edit your claude_desktop_config.json file. To find it, click on your name and then “Settings”. Next, click on “Developer” to see “Local MCP servers” and then click on “Edit Config”.

{
  "mcpServers": {
    "biocontext_kb": {
      "command": "uvx", // On Mac, use "brew install uv"
      "args": [
        "biocontext_kb@latest"
      ],
      "env": {
        "UV_PYTHON": "3.12"
      }
    }
  }
}

[!WARNING] Don't forget to restart Claude to apply the changes.

2. Local setup with uv

Run the server with streamable HTTP and uvicorn:

export MCP_ENVIRONMENT=PRODUCTION && export PORT=8000 && uvx biocontext_kb

Run the server with stdio transport:

export MCP_ENVIRONMENT=DEVELOPMENT && uvx biocontext_kb

3. Local setup with IDEs

Change the configuration file of your coding agents, e.g., VS Code (.vscode/mcp.json), Cursor (.cursor/mcp.json), or WindSurf (.codeium/windsurf/mcp_config.json):

{
  // VS Code: Use "servers" instead of "mcpServers"
  "mcpServers": {
    "biocontext_kb": {
      // if it doesn't work, replace with the `which uvx` path (installation via Homebrew recommended on macOS)
      "command": "uvx",
      "args": [
        "biocontext_kb@latest"
      ],
    }
  }
}

When using Windows and WSL2 the above config needs to be adapted as follows:

{
  // VS Code: "servers"
  "mcpServers": {
    "biocontext_kb": {
      "command": "wsl",
      "args": [
        "~/.local/bin/uvx", // set to the path to your `uvx`
        "biocontext_kb"
      ]
    }
  }
}

4. Docker

Clone the latest version of this repository:

git clone https://github.com/biocontext-ai/knowlegebase-mcp.git
cd knowlegebase-mcp

Then build the container, running gunicorn with multiple uvicorn workers:

# Build the docker container
docker build -t biocontext_kb:latest .
docker run -p 127.0.0.1:8000:8000 biocontext_kb:latest

This exposes your MCP server at: http://127.0.0.1:8000/mcp/

[!WARNING] For public deployments, you should disable unnecessary ports and access your MCP server through a reverse proxy, e.g., Nginx or Caddy. You may also want to configure the running user and the directory to have limited rights, use Docker or podman in a rootless setup and take additional security measures like DDOS protection with Cloudflare or fail2ban.

MCP Clients

To develop your own agentic systems that make use of the MCP server, you may want to consider some of the following options:

• PydanticAI
• FastMCP
• Langchain MCP adapters
• mcp-use

Publication & Documentation

You can find our Nature Biotechnology correspondence here: https://www.nature.com/articles/s41587-025-02900-9.

If our work is useful to your research, please cite it as below.

@article{BioContext_AI_Kuehl_Schaub_2025,
  title={BioContextAI is a community hub for agentic biomedical systems},
  url={http://dx.doi.org/10.1038/s41587-025-02900-9},
  urldate = {2025-11-06},
  doi={10.1038/s41587-025-02900-9},
  year = {2025},
  month = nov,
  journal={Nature Biotechnology},
  publisher={Springer Science and Business Media LLC},
  author={Kuehl, Malte and Schaub, Darius P. and Carli, Francesco and Heumos, Lukas and Hellmig, Malte and Fernández-Zapata, Camila and Kaiser, Nico and Schaul, Jonathan and Kulaga, Anton and Usanov, Nikolay and Koutrouli, Mikaela and Ergen, Can and Palla, Giovanni and Krebs, Christian F. and Panzer, Ulf and Bonn, Stefan and Lobentanzer, Sebastian and Saez-Rodriguez, Julio and Puelles, Victor G.},
  year={2025},
  month=nov,
  language={en},
}

Further Resources

• Project documentation: https://biocontext.ai
• API documentationt: https://docs.kb.biocontext.ai/
• BioContextAI Registry: https://github.com/biocontext-ai/registry
• Chat Interface: https://biocontext.ai/chat

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

[!WARNING] The Apache 2.0 License only applies to the code provided in this repository. For usage limitations and licenses of the individually integrated APIs, users should directly refer to the respective API providers. We provide an overview below.

Data Sources and Licensing

| Data Source

---

README truncated. View full README on GitHub.