Skip to main content

GeneBe MCP β€” let your AI assistant interpret variants with GeneBe

Piotr
GeneBe maintainer

TL;DR: GeneBe MCP is live. Add https://mcp.genebe.net/v1/mcp to Claude, ChatGPT, Cursor or VS Code, and your AI assistant can annotate and classify human genetic variants with GeneBe: an ACMG verdict with points and criteria, gnomAD frequencies, ClinVar, predictor scores and transcript consequences. It is free, read-only and needs no account.

AI assistants are good at explaining genetics and bad at remembering the status of one specific variant. They sound sure of themselves and are sometimes wrong. I wanted to fix that, so GeneBe now speaks MCP, the protocol that lets assistants call external tools.

Ask your assistant "is NM_005577.4:c.4195A>C benign?" or "what is chr6-160585140-T-G, and why is it classified that way?". Instead of guessing, it queries GeneBe and answers from the data, with the reasoning behind every ACMG criterion and a link back to the variant page.

What it does​

  • accepts coordinates, HGVS, rsID or SPDI,
  • returns the transparent, points-based ACMG classification together with its criteria,
  • adds gnomAD frequencies, predictor scores, ClinVar and per-transcript consequences,
  • converts between notations and genome builds (hg19, hg38, T2T),
  • shows gene details, including MANE Select and ClinGen gene–disease validity.

How to connect​

No registration, no API key. Add one URL to your assistant:

https://mcp.genebe.net/v1/mcp

For example, in Claude Code:

claude mcp add --transport http genebe https://mcp.genebe.net/v1/mcp

Step-by-step instructions for Claude, ChatGPT, Cursor and VS Code are in the documentation.

What it does not do (yet)​

The server is read-only and answers questions about single variants. Please do not send patient data to it, and remember that GeneBe is for research use only: classifications need review by a qualified specialist.

What is next​

I would like the assistant to work with your own data, too. The plan includes signing in, viewing and editing your samples, searching for a variant across your own samples, uploading samples and supervising computations. None of this is available yet, and the order may change. Tell me what would be most useful to you.

The code is open source: github.com/genebe-net/genebe-mcp. Try it, and let me know what to improve πŸ™