Skip to content

LLM Integration Guide

AI coding assistants can accelerate BioCypher development.
These guides help LLMs understand BioCypher's patterns and conventions.

The easiest way to integrate specific BioCypher instructions into your AI-assisted workflow is to connect your software to our dedicated MCP server at https://mcp.biocypher.org/mcp (not human-readable). The copilot can help you use our cookiecutter template to get started (this one is also human-usable).

Note

This functionality is currently experimental and does not cover all BioCypher functionality yet. Use with caution.

This section provides specialized documentation for AI coding assistants (Copilot, Cursor, Claude, etc.) to help them understand BioCypher's architecture and create high-quality code that follows our conventions. For instance, for Cursor or VSCode, add it like this:

{
  "mcpServers": {
    "biocypher-mcp": {
      "url": "https://mcp.biocypher.org/mcp",
      "transport": "http"
    }
  }
}

You can then ask for guidance in building your knowledge representation. For instance:

Using the CSVs in the `data` directory, how do I build a BioCypher knowledge graph?

Further LLM-Specific Documentation and Information

BioCypher follows specific patterns and conventions that may not be immediately obvious to AI assistants. These guides provide:

  • Clear architectural patterns for common tasks
  • Concrete examples with expected input/output formats
  • Validation rules to ensure code quality
  • Best practices specific to BioCypher's domain

Available Guides (deprecated, use the MCP instead)

For AI assistants, the following .txt files are available in the root of this documentation:

How to Use These Guides

When working with an AI assistant on BioCypher code:

  1. Reference the specific guide for your task (e.g., adapter creation)
  2. Provide the guide content in your prompt for context
  3. Specify the schema configuration you're working with
  4. Request validation against the patterns described

Example Prompt

I'm creating a BioCypher adapter for NCBI GEO data. Please follow the adapter creation guide at llms-adapters.txt.

My schema configuration defines these node types:
- geo_sample (input_label: "geo_sample")
- geo_series (input_label: "geo_series")

And these edge types:
- HAS_SAMPLE (input_label: "HAS_SAMPLE")

Please create an adapter that follows the 3-tuple (node_id, node_label, attributes_dict) and 5-tuple (edge_id, source_id, target_id, edge_label, attributes_dict) patterns described in the guide.