gi-mcp

Research and development use. Not for clinical or diagnostic decisions. For high-volume or latency-sensitive use, contact contact@genomicintelligence.ai to provision a production-shaped deployment.

An MCP server for the Genomic Intelligence DNA-analysis API. It exposes the six inference tasks (promoter, splice, enhancer, chromatin, expression, annotation), a composite annotation→expression workflow, sequence acquisition (Ensembl + local FASTA), reference resources, and ready-made prompt workflows — all over MCP stdio.

It runs locally as a thin protocol translator: it owns no inference and stores no key, forwarding each request to your Genomic Intelligence backend under your API key.

Install & configure

gi-mcp runs with uv (uvx fetches and runs it with no manual install):

uvx gi-mcp                # recommended
pip install gi-mcp        # alternative

Request a partner key from https://genomicintelligence.ai/contact, then point your MCP client at uvx gi-mcp with GI_API_KEY in the environment. For Claude Desktop, edit claude_desktop_config.json and reopen:

{
  "mcpServers": {
    "genomic-intelligence": {
      "command": "uvx",
      "args": ["gi-mcp"],
      "env": { "GI_API_KEY": "gi_..." }
    }
  }
}

For Claude Code: claude mcp add genomic-intelligence --env GI_API_KEY=gi_... -- uvx gi-mcp.

Upgrading from 0.1.0a17 or earlier

Pinned below 0.1.0a18? Upgrade. That release fixed a coordinate bug, not a cosmetic one: fetch_gene_for_expression placed the TSS of a minus-strand gene at gene-sense offset 4,598 instead of 4,599, so the 9,198 bp expression window was shifted one base. A shifted window does not fail — it returns a confident number for the wrong sequence. Plus-strand genes were unaffected.

uvx gi-mcp already tracks the newest release; pip install -U gi-mcp if you installed with pip, and re-pin if you pin (uvx gi-mcp@<version>).

📖 Full documentation

https://docs.genomicintelligence.ai/mcp.md — per-client setup, the handle pattern, copy-paste "try it" prompts, the full tool/resource surface, and the underlying /v1 contract. Start there.

Environment

Where reference data comes from

Gene symbols resolve against a bundled coordinate catalog (109 pre-resolved loci across human, mouse, rat, zebrafish, fly, yeast and arabidopsis) before anything reaches rest.ensembl.org; anything outside it still resolves live. Sequence bytes come from a response cache, then UCSC, then Ensembl, and the bases behind the published example prompts ship in the package so they need no network at all. When every upstream is unreachable a cached copy is served rather than an error, and the response's meta says so (source, locus_source, assembly, catalog_release, stale). Reference sequence for a fixed assembly does not change, so a cached answer is the same bases a live fetch would have returned — verified byte-identical between UCSC and Ensembl on every control locus.

Development & contributing

Setup, conventions, the architecture walkthrough, and the release process live in CONTRIBUTING.md. Quick start:

python3 -m venv .venv && .venv/bin/pip install -e ".[dev]"
.venv/bin/pytest                 # full suite (mocked; no network)

Release history is in CHANGELOG.md.