LABSCO SUMMARY

The assembly restriction is a feature.

Only GRCh38/hg38 coordinates are accepted, so a variant given in an older assembly fails loudly instead of returning confident annotations for the wrong position — the failure mode that quietly ruins analyses. The other detail worth planning around: the annotator set is a curated default, so check what is actually running before treating an absent annotation as a negative result.

What it is

A remote server that puts OpenCRAVAT variant annotation behind MCP tools. You give it a variant in whichever notation you have, and it returns annotations from a curated default set of OpenCRAVAT annotators.

What you get

Annotation from four kinds of input — genomic coordinates as chrom/pos/ref/alt, a dbSNP rsID, a ClinGen Allele Registry ID, or HGVS in g., c. or p. form
Discovery of which OpenCRAVAT annotators are being run, so you know what the answer is based on
Inspection of an annotator's output schema — field names, types and descriptions — before you parse its results
Conversion of a protein missense notation such as BRAF V600E into candidate genomic HGVS changes via SynVar, which you can then annotate
Coordinates are GRCh38/hg38 only — other assemblies are rejected rather than silently mis-annotated

Requirements

Nothing to install: it is hosted at https://mcp.opencravat.org/mcp, and the documented setup explicitly says to add it with no auth — no user-specific data is needed. In ChatGPT it goes in as a custom app with authentication set to "No Auth". To run it locally instead, it is a Cloudflare Worker: npm install then npm start puts the endpoint at http://localhost:8787/mcp, reachable by running mcp-remote under npx. Node 18+ is recommended for the local path.

Setup effort

Paste a URL — add the endpoint to your client

See installation options

USE CASES

01

Annotating a variant you have as an rsID

Pass the rsID directly rather than converting to coordinates first.

02

Starting from a protein change

Convert something like BRAF V600E into candidate genomic HGVS, then annotate those.

03

Knowing what the annotation actually covers

List the annotators in the default set and read their output schemas before interpreting results.

04

Getting variant context into a chat without local setup

Connect to the hosted endpoint with no authentication and no install.

INSTALLATION

Connect your client

Server setup

Setup

TERMINAL

git clone https://github.com/KarchinLab/oc-mcp
cd oc-mcp
npm install
npm start

Hosted endpoint

ENDPOINT URL

https://mcp.opencravat.org/mcp

Authentication: None — add the URL with no authentication; no user-specific data is needed.

MCP Clients

Claude Desktop — local

CLAUDE DESKTOP — LOCAL · JSON

{
  "mcpServers": {
    "OpenCRAVAT": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "http://localhost:8787/mcp"
      ]
    }
  }
}

Claude.ai

Add it as a custom connector, following Claude's guide: https://support.claude.com/en/articles/11175166-getting-started-with-custom-connectors-using-remote-mcp

Use the URL https://mcp.opencravat.org/mcp and do not set up auth.

CLAUDE.AI · URL

https://mcp.opencravat.org/mcp

Enable the connector for your chat.

ChatGPT

Open Settings → Apps → Advanced Settings and turn on Developer Mode.

Click Create App and use https://mcp.opencravat.org/mcp as the MCP Server URL.

Set Authentication to No Auth and give the app a name, such as OpenCRAVAT.

In a new chat, add the app with the + icon in the lower left of the chat box.

OpenCRAVAT