Plant Genomics MCP Server

🌱 plant-genomics-mcp

56 tools for plant-genomics locus lookup over the Model Context Protocol — 29 single-locus + 1 motif lookup + 1 region query + 1 assembly listing + 1 release report + 1 variant annotator + 1 gene-set enrichment + 1 BLAST search + 2 member lists (family entry, gene tree) + 13 parallel-batch + 5 cross-source synthesis variants. 23 free, public sources: Ensembl Plants, Phytozome BioMart, UniProtKB, Europe PMC, QuickGO, Planteome, PlantCyc/PMN, g:Profiler, AlphaFold DB, PDBe, InterPro, JASPAR, PANTHER, OrthoDB, AraGWAS, 1001 Genomes, NCBI BLAST, Gramene, KEGG, STRING-DB, ATTED-II, ThaleMine, and BAR (Bio-Analytic Resource for Plant Biology).

📦 Install

# Zero-install — uv fetches and runs it on demand
claude mcp add plant-genomics --scope local -- uvx plant-genomics-mcp

# pipx — installs the CLI onto your PATH
pipx install plant-genomics-mcp
claude mcp add plant-genomics --scope local -- plant-genomics-mcp

# GHCR Docker image
docker pull ghcr.io/musharna/plant-genomics-mcp:latest
claude mcp add plant-genomics --scope local -- \
  docker run --rm -i ghcr.io/musharna/plant-genomics-mcp:latest

# From source
git clone https://github.com/musharna/plant-genomics-mcp.git
cd plant-genomics-mcp
python -m venv .venv && .venv/bin/pip install -e .
claude mcp add plant-genomics --scope local -- "$(pwd)/.venv/bin/plant-genomics-mcp"

💬 Try it

Once connected, ask Claude a plain-language question — you don't have to name any tool or remember the chain:

"Tell me everything about the Arabidopsis gene AT1G01010 — its function, GO terms, KEGG pathways, protein-interaction partners, and recent papers."

The server supplies the tools (here: Ensembl Plants, UniProt, QuickGO, KEGG, STRING-DB and Europe PMC lookups); which ones get called, in what order and in how many turns is up to the client. In the recording at the top of this page (a narrower prompt), Claude Code picked the calls itself and returned one combined answer. Swap in any locus and pass organism= for cross-species — e.g. rice Os01g0100100 (oryza_sativa) — and each tool maps the organism to that backend's own identifier.

A worked 64-call run over 114 genes in three organisms, with the 51 gaps it logged (39 since closed), is in examples/arf_family/PAGE.md.

🛠️ Tools

56 tools across 23 backends — Ensembl Plants, Phytozome BioMart, UniProtKB, Europe PMC, QuickGO, Planteome, PlantCyc/PMN, g:Profiler, AlphaFold DB, PDBe, InterPro, JASPAR, PANTHER, OrthoDB, AraGWAS, 1001 Genomes, NCBI BLAST, Gramene, KEGG, STRING-DB, ATTED-II, ThaleMine, BAR. 29 single-locus + 1 motif lookup + 1 region query + 1 assembly listing + 1 release report + 1 variant annotator + 1 gene-set enrichment + 1 BLAST search + 2 member lists + 13 parallel-batch + 5 cross-source synthesis. Most take a TAIR-style locus (e.g. AT1G01010) plus optional organism= (slug / scientific name / common name / NCBI taxid — 12-plant curated coverage matrix at the pgmcp://organisms/coverage MCP resource). All publish JSON outputSchema, EDAM ontology tags, and behaviour annotations — every tool is readOnlyHint + openWorldHint, so hosts can surface them without a destructive-action confirmation prompt.

⚡ Quickstart

After install, the simplest call returns the Ensembl Plants record for NAC001 — the canonical worked example used throughout examples/:

// arguments
{ "locus": "AT1G01010" }

// result (truncated)
{
  "id": "AT1G01010",
  "organism": "arabidopsis_thaliana",
  "display_name": "NAC001",
  "biotype": "protein_coding",
  "seq_region_name": "1",
  "start": 3631,
  "end": 5899,
  "strand": 1,
  "assembly_name": "TAIR10",
  "description": "NAC domain containing protein 1 ..."
}

Cross-species — pass organism=:

{ "locus": "Os01g0100100", "organism": "oryza_sativa" }

A recorded Claude Code session (2026-05-24) with a narrower prompt — the Ensembl record, UniProtKB entry and top three Europe PMC papers for AT1G01010 — answered it in one turn:

Full per-tool walkthroughs (with real upstream-API transcripts) live in examples/:

📚 Resources & prompts

Clients discover them via resources/list and prompts/list.

Resources (resources/read):

Prompts (prompts/get):

🔌 Transports

The HTTP transport is stateless and emits JSON responses by default — the right shape for registry indexers and remote hosting.

Self-hosting

There is no public hosted endpoint. To use the HTTP transport, run it yourself: it is the same binary, gated by your own bearer token (PLANT_GENOMICS_MCP_HTTP_TOKEN), with NCBI BLAST requests sent under your own contact email.

⚙️ Configuration

Stdio needs no configuration. The two env vars that matter:

The cache is process-local — restart the server to drop all entries. Long-running calls (retry storms, multi-second Phytozome BioMart POSTs) emit MCP notifications/progress over the active session; clients opt in via progressToken in the request _meta.

⚠️ Error model

All live tools raise PlantGenomicsError subclasses; the MCP SDK stringifies them into the wire content with a [ClassName] prefix so clients can route on failure kind without parsing the message:

Batch tools return {tool, count, results, errors} where results[locus] is the same shape as the single-locus tool and errors[locus] is the same [ClassName] message string. Ensembl's batch uses the native POST /lookup/id endpoint (one HTTP round-trip); everything else fans out via asyncio.gather.

🧪 Development

.venv/bin/pip install -e '.[dev]'                         # or: uv sync --extra dev
.venv/bin/pytest -q                                       # unit tests
PLANT_GENOMICS_MCP_LIVE=1 .venv/bin/pytest -q             # adds live network probes
PLANT_GENOMICS_MCP_STDIO_SMOKE=1 .venv/bin/pytest -q      # adds stdio smoke
.venv/bin/ruff check .

With uv, pass --extra dev — a bare uv sync omits (and removes) the test dependencies. See CONTRIBUTING.md.

CI runs the unit suite + the stdio smoke on every push/PR (matrix: Python 3.11, 3.12, 3.13, 3.14 — the full requires-python range), with a dead proxy set so an ungated network call fails. A separate live-smoke job runs two live test files (tests/test_verify_genes.py, tests/test_arf_mcp_client.py, InterPro and PANTHER) on Python 3.12 on the same pushes and PRs; it reports a result but is not a required check, because its result depends on third-party services. The rest of the live suite (PLANT_GENOMICS_MCP_LIVE=1) is not run in CI.

Drift detection. scripts/benchmark_annotations.py runs a curated corpus of 27 loci (spanning all 12 organisms) through 12 functions — the organism resolver plus 11 lookups across 9 of the 23 backends (ATTED-II, BAR, Ensembl Plants, Europe PMC, Gramene, KEGG, Phytozome, STRING-DB, UniProt) — and compares the results to a frozen snapshot of earlier results (scripts/benchmark_annotations.expected.json), emitting PASS / DRIFT / FAIL plus two cross-source consistency invariants. A change is detected relative to that snapshot, not checked against an independent truth. BLAST and the synthesis pipelines are registered in the script, but no corpus locus exercises them. A scheduled GitHub Actions workflow (.github/workflows/benchmark.yml) runs it weekly and pages when the same loci fail on a re-run. Operator guide: docs/benchmarking.md.

.venv/bin/python scripts/benchmark_annotations.py        # full live sweep

See CHANGELOG.md for release notes, including the v0.8 → v0.9 species=/organism_id= → organism= migration and the v1.0.1 HTTP-token enforcement change.

MCP registry

Listed in the official MCP registry under the namespace below (ownership-verification token for mcp-publisher):

mcp-name: io.github.musharna/plant-genomics-mcp

License

MIT — see LICENSE. Underlying services (Ensembl Plants, Phytozome, TAIR, PlantCyc, BAR) have their own terms of use; consult each before bulk querying.