STRING MCP Server

STRING provides a Model Context Protocol (MCP) server that allows AI agents and other MCP-compatible tools to discover and interact with the STRING database. The MCP acts as a standardized bridge, describing what STRING tools and data are available and how they can be used.

Unlike a simple API wrapper, the STRING MCP server is purpose-built for AI agents. Each tool is described with metadata, hints, and examples specifically designed to help large language models use the tools correctly and interpret the resulting data. The MCP layer curates what and how data are returned, with a focus on context size and readability.

You can either host the server yourself (and modify it as needed) or use the remote server provided by STRING. For non-AI integration, it is recommended to use the STRING API directly. If you need complete datasets, you should use the download page.

Using the STRING MCP Server Top ↑

A remote MCP server is available at:

https://mcp.string-db.org/

This server is compatible with major AI agents (e.g. OpenAI custom GPTs, Anthropic Claude).

It provides access to a set of protein network analysis tools, including:

  • Protein identifier resolution
  • Interaction network data and figures
  • Functional enrichment results and figures
  • Functional annotation
  • Homology search
  • Links to interactive evidence pages

Every MCP tool definition includes tailored documentation (docstrings) optimized for large-language-model consumption. These definitions make the MCP interface self-describing, so an AI agent can autonomously discover, call, and interpret STRING functionality without requiring detailed knowledge of the database.

For the full list of tools, see the server code.

For non-AI usage, we recommend the STRING API, which is designed for full reproducibility and programmatic integration.

STRING Chat Top ↑

STRING Chat is a natural-language interface that directly uses the provided MCP server. It serves as a working example of how AI agents can access and combine STRING data in real time using structured tool definitions exposed by the MCP.

It supports natural language queries such as:

  • “Show me the interaction network for TP53.”
  • “What are the homologs of BRCA1 in zebrafish?”
  • “Which pathways are enriched in this list of proteins?”

Use cases:
STRING Chat is best suited for quick exploration, conversational queries. It can invoke multiple MCP tools in sequence, aggregate their outputs, and summarize the combined results into a unified response. The interface supports all major languages, and example questions and use cases are shown directly below the chat input.

Important notes:

  • The chat is not a replacement for the main STRING web interface.
  • For large-scale or precise queries, use the STRING web interface or the API to ensure reproducibility and access to the full dataset.
  • AI-generated responses may mix STRING results with inferred or external knowledge.
  • Always request direct STRING links to verify and reproduce results.

Versioning Top ↑

https://mcp.string-db.org will always provide the newest version of the server.
The server is still under testing and development, so certain aspects of its behavior may change.
A versioned address will be released soon after the public release.

Run It Yourself Top ↑

The source code for the Python implementation of the STRING MCP server is available here:

GitHub code repository

Running locally allows you to:

  • Modify or extend it with your own tools
  • Customize the behavior of tools
  • Add data sources other than STRING

The implementation is licensed under The MIT License. The output data is licensed under CC-BY-4.0.