Summary
A comprehensive genomics toolkit that wires Claude into 11 plant biology databases: Ensembl Plants, Phytozome, UniProtKB, Europe PMC, QuickGO, NCBI BLAST, Gramene, KEGG, STRING-DB, ATTED-II, and BAR. You pass it a locus identifier like AT1G01010 and get back gene metadata, cross-references, protein records, literature mentions, GO annotations, homology trees, pathway memberships, protein interactions, coexpression neighbors, and tissue-specific expression profiles. The 32 tools split into single lookups, parallel batch operations, and cross-source synthesis calls that merge data from multiple backends. Reach for this when you're doing comparative plant genomics, pathway analysis, or need to pull together fragmented annotations across disparate public repositories without writing custom scrapers for each one.
Keeps your Mac awake while Claude Code, Codex or Cursor works. Lets it sleep when they are done.
Try free for 7 days →
Block distracting apps from your iPhone permanently without a 3rd party app. Free and open source.
Block now (100% free) →
Notes, actions and memory. Without a meeting bot. First month 100% off.
Download for free →
belt cli automatically finds the best tools and skills for your agent. image, video, music, tts...
one prompt install →
Free an open-source skills that make AI coding agents easier to understand, verify, and control.
Download for free →
Connect your Claude agent to live crypto prices and trading routes via 1inch
Get the MCP →🌱 plant-genomics-mcp
Plant gene lookup over the Model Context Protocol. Give Claude, or any MCP client, a locus and it can query 23 free, public databases, from Ensembl Plants and UniProt to KEGG, STRING and AlphaFold, with curated coverage of 12 plants.
Method, timings and caveats: Does it help?
📦 Install
# Zero-install — uv fetches and runs it on demand
claude mcp add plant-genomics --scope local -- uvx plant-genomics-mcp
Other install paths (pipx, Docker, from source)
# 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
Ask in plain language; you don't name the tools or chain them:
"Tell me everything about the Arabidopsis gene AT1G01010 — its function, GO terms, KEGG pathways, protein-interaction partners, and recent papers."
The client picks which tools to call. For another species, name it or pass
organism= (e.g. rice Os01g0100100, oryza_sativa).
Below are two recorded runs with narrower prompts: NAC001's Ensembl record and
UniProt entry, then its rice and maize orthologs, where Claude checked three
sources and said which one does not back the rice match. Both prompts and full
answers, including the first run's papers: examples/assets/cards/RUN.md.
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.
Per-tool walkthroughs with real upstream transcripts, including the one-call gene
dossier gene_report_AT1G01010.md, are in
examples/.
📊 Does it help?
We asked Claude Opus 5.5, in Claude Code, 52 plant gene-lookup questions (location, protein length, name → locus, UniProt accession, ortholog; 12 genomes), three times each. Alone, it declined most of them. With web search it got 48 of the 52 right in every run, mostly by calling the Ensembl and UniProt REST APIs itself. With this server it got all 52 right in every run, and its slowest answer took 86 s, against 705 s with web search. The answer key comes from the databases the server reads, so this measures whether lookups are done correctly. The chart at the top shows every answer. Cost, timings and caveats: benchmarks/agent_qa/.
🛠️ 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 and BAR. That is 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 locus (e.g. AT1G01010) and an optional organism=: a slug,
scientific or common name, or NCBI taxid (the 12-plant coverage matrix is the
pgmcp://organisms/coverage resource). Every tool is marked read-only and
publishes a JSON outputSchema.
Full tool matrix
| # | Category | Tool | What it does |
|---|---|---|---|
| 1 | Gene metadata (live) | ensembl_plants_lookup_locus | Fetches gene record from Ensembl Plants REST (any plant species). |
| 2 | Cross-references (live) | get_gene_xrefs | Fetches cross-DB references (UniProt, NCBI Gene, TAIR, GO, …) from Ensembl. |
| 3 | Gene metadata (live) | phytozome_lookup_locus | Fetches gene record from Phytozome BioMart (any Phytozome proteome). |
| 4 | Protein (live) | resolve_locus_to_uniprot | Resolves a locus to its UniProtKB record (Swiss-Prot preferred, TrEMBL OK). |
| 5 | Literature (live) | locus_literature | Searches Europe PMC for papers mentioning the locus (free, no API key). |
| 6 | GO annotations (live) | locus_go_annotations | Fetches QuickGO GO annotations (locus → UniProt → QuickGO). |
| 7 | Sequence search (live) | blast_sequence | NCBI BLAST URLAPI — async Put/Get polling with progress notifications. |
| 8 | Homology (live) | gramene_homologs | Fetches Gramene v69 homology entries (ortholog / paralog) with gene_tree_id. |
| 9 | Pathways (live) | kegg_pathways | Fetches KEGG pathway memberships. 7 organisms: Arabidopsis (ath:, native AGI), + rice (osa:), maize (zma:), soybean (gmx:), barley (hvg:), poplar (pop:), brachypodium (bdi:) bridged via Ensembl → Entrez ID. |
| 10 | Interactions (live) | string_interactions | Fetches STRING-DB first-neighbor interaction partners with per-channel score. |
| 11 | Coexpression (live) | atted_coexpression | Fetches ATTED-II top-N coexpression neighbors with a score: z-score for Arabidopsis, logit score (LSmr) for the other releases. |
| 12 | Curator summary (live) | bar_gene_summary | Fetches BAR ThaleMine + GAIA-aliases curator summary for an Arabidopsis locus. |
| 13 | Expression (live) | bar_efp_expression | Fetches BAR eFP-Browser expression profile (mean ± SD per tissue) for a locus. |
| 14 | Interactions (live) | bar_aiv_interactions | Fetches BAR AIV interaction partners (Arabidopsis + rice) with confidence + papers. |
| 15 | Curator summary (live) | tair_locus_info | Silent upgrade — alias of bar_gene_summary. MCP tool name preserved for clients. |
| 16 | Metabolism (live) | plantcyc_locus_info | Walks gene → enzyme → reactions → PlantCyc/PMN pathways (free BioCyc web-services API). The metabolic-pathway view KEGG/GO lack; found=false for non-enzymatic genes. 11 species have a PGDB. |
| 17 | Sequence (live) | get_sequence | Fetches a locus's sequence (genomic / cds / cdna / protein) from Ensembl /sequence/id — the fetch half of lookup → fetch → BLAST; feed sequence to blast_sequence. |
| 18 | Region query (live) | ensembl_region_query | Lists gene/transcript/cds/exon features overlapping a genomic interval (chr:start-end) via Ensembl /overlap/region — "what's in this QTL interval" without a per-locus lookup. |
| 19 | Enrichment (live) | go_enrichment | GO + KEGG over-representation for a gene list via g:Profiler g:GOSt — "what is my DE / co-expression set enriched for?" Reports unmapped loci; optional custom background. All 12 organisms. |
| 20 | Plant ontology (live) | locus_plant_ontology | Plant Ontology (anatomy / dev-stage) + Trait Ontology annotations for a locus via Planteome (Solr) — the plant-specific ontologies GO doesn't cover. by_ontology rollup; taxon-filtered. Strong for 6 species. |
| 21 | Structure (live) | alphafold_structure | AlphaFold DB predicted 3D model for a locus (locus → UniProt → model): global mean pLDDT, per-band confidence, modelled span, and mmCIF / PDB / PAE URLs. found=false when no model is deposited. All 12 organisms. |
| 22 | Structure (live) | experimental_structures | PDBe experimentally-solved (X-ray / cryo-EM / NMR) structures for a locus (locus → UniProt): best-first PDB id, chain, method, resolution, coverage, residue span. found=false when none deposited (common for plants). All 12 organisms. |
| 23 | Domains (live) | interpro_domains | InterPro domain / family architecture (locus → UniProt): each entry's accession, name, type, source_database (Pfam included), integrated InterPro id, and residue spans, plus a count_by_type rollup. All 12 organisms. |
| 24 | TF motifs (live) | tf_binding_motifs | JASPAR curated TF DNA-binding profiles for a locus (locus → UniProt → symbol search, then UniProt-confirmed): matrix id, TF class/family, assay type (SELEX / ChIP-seq / PBM / DAP-seq), IUPAC consensus, PubMed refs, logo URL. Fuzzy name hits for other genes are quarantined in name_only_matches. Arabidopsis-heavy coverage. |
| 25 | TF motifs (live) | jaspar_motif | One JASPAR profile by matrix id (e.g. MA0570.1, or MA0570 for the newest version) including the raw position-frequency matrix — the drill-down companion to tf_binding_motifs. |
| 26 | Interactions (live) | experimental_interactions | ThaleMine CURATED EXPERIMENTAL interaction partners (BioGRID / IntAct / PSI-MI) for an Arabidopsis locus — per partner: detection method (two hybrid, pull down, ...), PSI-MI relationship type, physical vs genetic, source DB, PubMed IDs, and an evidence count. The experimental counterpart to string_interactions (predicted / text-mined). Arabidopsis only. |
| 27 | Function (live) | locus_gene_rifs | ThaleMine curated GeneRIF statements — one-sentence, manually curated descriptions of what the gene does, each tied to a PubMed ID (HY5 has 114). Citable functional context that GO terms and raw abstracts don't provide. Arabidopsis only. |
| 28 | Variation (live) | locus_variants | Natural (EVA/dbSNP) variants overlapping a locus's genomic span via Ensembl /overlap/region — id, source, consequence class, alleles, clinical significance. variant_count + truncated. All 12 organisms. |
| 29 | Variation (live) | vep_annotate | Ensembl VEP consequence prediction for a variant (region + allele, not locus) — most-severe consequence + per-transcript SO terms, IMPACT, SIFT. All 12 organisms. |
| 30 | Orthology (live) | panther_family | PANTHER protein family + subfamily (id + name), GO terms by aspect, protein class, and pathways. found=false when unmapped. All 12 organisms. |
| 31 | Orthology (live) | orthodb_orthologs | OrthoDB ortholog group (name, evolutionary rate) + cross-species member genes at the Viridiplantae level. organism_count + truncated. All 12 organisms. |
| 32 | Diversity (live) | aragwas_associations | AraGWAS genome-wide association hits per locus — score, MAF, SNP effect, phenotype/study. Arabidopsis-only. |
| 33 | Diversity (live) | arabidopsis_natural_variation | 1001 Genomes natural-variation SNP effects across 1135 accessions — chr, position, effect, impact, amino-acid change, transcript + gene span. Arabidopsis-only. |
| 34 | Batch (live) | batch_* (twelve dedicated variants) | Parallel per-locus fanout for tools 1–6, 8–12, 14. Up to 50 loci per call. |
| 35 | Batch (live) | batch_locus_call | Runs any tool whose only required argument is locus (33 tools, including gene_report and the synthesis tools) over up to 50 loci; the shared args are checked against that tool's schema once, before any call. |
| 36 | Synthesis (live) | *_synth / consensus_homologs (four) | Compose 2–5 backends in parallel, return a SynthesisEnvelope with per-step status. |
| 37 | Synthesis (live) | gene_report | One-shot "tell me about this gene" dossier — annotation + xrefs + protein + domains + GO + KEGG + STRING + literature composed into a rendered Markdown result.markdown (+ structured result.sections). |
| 38 | Families (live) | entry_members | Every protein in one organism carrying an InterPro / Pfam / PANTHER entry, with the locus each maps to (entry → genes; the reverse of tools 23 and 30). UniProt total + cursor paging; reviewed-only by default. |
| 39 | Families (live) | gene_tree_members | Every gene in the Ensembl Compara gene tree a locus is in now (or a gene_tree_id), with locus, protein id, species and organism; target_organism keeps one organism's members. total + limit. |
| 40 | Homology (live) | ensembl_plants_paralogs | Paralogues Ensembl Compara records for a locus — within_species_paralog and the other_paralog ("ancient paralogues") that gramene_homologs drops — closest first, with perc_id, taxonomy level and protein id. Not a family list: test membership with interpro_domains. total + limit. |
| 41 | Region query (live) | ensembl_plants_assembly | An organism's Ensembl assembly — name, GCA accession, karyotype — and every top-level seq-region with its length: the region names ensembl_region_query takes, chromosomes first, so a region walk is planned before its first call. total + limit. |
| 42 | Provenance (live) | upstream_release | The release a backend's own endpoint calls current (Ensembl, STRING, QuickGO, JASPAR, KEGG), for the backends whose answers state none; PDBe, AraGWAS and Europe PMC publish none and say why. A separate request, so read it before and after a run: equal values mean no release changed. |
📚 Resources & prompts
Four read-only MCP resources + three parameterized prompts
Clients discover them via resources/list and prompts/list.
Resources (resources/read):
| URI | What |
|---|---|
pgmcp://cache/stats | Per-backend TTLCache rollup — {hits, misses, size} for each live backend. |
pgmcp://organisms/phytozome | Slug → Phytozome organism_id map. |
pgmcp://backends/status | Static per-backend roster — name, base_url, citation DOI, subscription_gated, kind. Nothing is probed: kind is always "live" and says nothing about whether the backend is up right now. |
pgmcp://organisms/coverage | Markdown table of all 12 supported plants × 9 ID slots (ncbi_taxid / ensembl / phytozome / string / europe_pmc / kegg / atted / gprofiler / plantcyc). |
Prompts (prompts/get):
| Name | Required | Optional | Chains |
|---|---|---|---|
analyze_locus | locus | organism (default arabidopsis_thaliana) | Ensembl → xrefs → UniProt → Europe PMC → QuickGO. |
find_homologs | sequence | program (default blastp) | blast_sequence → per-hit resolve_locus_to_uniprot for UniProt-shaped accessions. |
biological_context | locus | top_n (default 10) | Gramene → KEGG → UniProt → STRING → ATTED-II. |
🔌 Transports and configuration
stdio (plant-genomics-mcp, the default) needs no configuration.
plant-genomics-mcp-http serves stateless streamable HTTP at
http://host:port/mcp. There is no public endpoint: you run it yourself, behind
your own token.
| Variable | When | Effect |
|---|---|---|
PLANT_GENOMICS_MCP_HTTP_TOKEN | HTTP transport only | Bearer token for /mcp; must be ≥32 chars or the HTTP server aborts at startup. Generate openssl rand -hex 32. |
PLANT_GENOMICS_MCP_NCBI_EMAIL | If you use BLAST | NCBI etiquette contact. Unset → placeholder + per-call warning; NCBI may throttle. |
All env vars (HTTP bind, body cap, cache, BLAST concurrency)
| Variable | Default | Effect |
|---|---|---|
PLANT_GENOMICS_MCP_HTTP_HOST | 127.0.0.1 | HTTP bind address. |
PLANT_GENOMICS_MCP_HTTP_PORT | 8765 | HTTP TCP port. |
PLANT_GENOMICS_MCP_HTTP_MAX_BODY | 2097152 (2 MiB) | Reject POSTs with Content-Length larger than this. |
PLANT_GENOMICS_MCP_HTTP_STATELESS | 1 | 0 keeps per-client session state (SSE-style). |
PLANT_GENOMICS_MCP_HTTP_JSON | 1 | 0 switches the response shape to streaming SSE events. |
PLANT_GENOMICS_MCP_BLAST_CONCURRENCY | 2 | Max in-flight BLAST searches per process (NCBI per-IP rate limit). |
PLANT_GENOMICS_MCP_CACHE_TTL | 600 | Per-backend TTL+LRU cache entry lifetime, in seconds. 200-only. |
PLANT_GENOMICS_MCP_CACHE_SIZE | 256 | Max entries per backend before LRU eviction. |
PLANT_GENOMICS_MCP_CACHE_DISABLED | unset | Any non-empty value makes every cache a no-op. |
The cache is per process; restart the server to clear it. Long calls (BLAST,
Phytozome BioMart) send MCP progress notifications when the request carries a
progressToken.
⚠️ Error model
Wire-prefix taxonomy + batch result shape
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:
| Wire prefix | When |
|---|---|
[NotFoundError] | 404 / empty BioMart row / invalid locus identifier |
[RateLimitError] | 429 retry budget exhausted — back off and retry |
[UpstreamUnavailableError] | 5xx past retry budget — service outage, try a peer backend |
[PlantGenomicsError] | Other (BioMart Query ERROR: body, unexpected column count, etc.) |
Batch tools return {tool, count, results, errors}: each results[locus] has
the single-locus shape and each errors[locus] is the same [ClassName] message
string.
🧪 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 .
A bare uv sync omits the test dependencies. CI runs the unit suite and the
stdio smoke on Python 3.11–3.14 with the network blocked, plus a non-required
live-smoke job against the real services. A weekly job checks 27 loci against
a frozen snapshot and pages on drift (docs/benchmarking.md).
Setup: CONTRIBUTING.md. Release notes:
CHANGELOG.md.
MCP registry
Listed in the official MCP registry;
mcp-publisher verifies ownership from this line:
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.
Featured
Keeps your Mac awake while Claude Code, Codex or Cursor works. Lets it sleep when they are done.
Try free for 7 days →Block distracting apps from your iPhone permanently without a 3rd party app. Free and open source.
Block now (100% free) →Notes, actions and memory. Without a meeting bot. First month 100% off.
Download for free →belt cli automatically finds the best tools and skills for your agent. image, video, music, tts...
one prompt install →Free an open-source skills that make AI coding agents easier to understand, verify, and control.
Download for free →Connect your Claude agent to live crypto prices and trading routes via 1inch
Get the MCP →