Plant Genomics MCP

What musharna/plant-genomics-mcp MCP server does

The musharna/plant-genomics-mcp MCP server gives an MCP-compatible agent read-only access to public plant-biology data services. It is designed around plant locus identifiers such as Arabidopsis TAIR IDs, but several tools also accept species slugs, scientific or common names, and NCBI taxids. Responses use JSON schemas, and the tools include EDAM ontology tags and read-only behavior metadata.

The server covers gene records, cross-references, genomic intervals, sequences, protein records, literature, GO and plant-specific ontology annotations, pathway membership, homology, protein interactions, expression, structures, domains, transcription-factor motifs, variants, genome-wide association results, and gene-set enrichment. It also includes batch and cross-source synthesis variants according to the project README.

How it works

Most calls begin with a locus and an optional organism. Ensembl Plants provides metadata, sequences, region features, and variation data; other tools resolve the locus through UniProt before querying services such as QuickGO, AlphaFold DB, PDBe, InterPro, STRING-DB, or JASPAR. Gene-list enrichment uses g:Profiler, while sequence searches use NCBI BLAST. Literature searches use Europe PMC, which does not require an API key.

The outputs are intended to be chained. For example, a sequence returned by get_sequence can be passed to the BLAST tool, and a UniProt resolution can provide the protein identifier needed for structure, domain, or interaction queries. Some tools report found=false for valid genes without a matching annotation or deposited structure; this is distinct from an unknown locus or unsupported organism.

Setup and configuration

Install the musharna/plant-genomics-mcp MCP server with uvx plant-genomics-mcp, or install the plant-genomics-mcp package with pipx and run its command-line entry point. The README also documents a Docker image and installation from source. The project requires Python 3.11 or newer and is licensed under MIT.

The documented sources are public and the listed tools do not require user API keys. Operation still depends on network access and the availability, identifiers, coverage, and request limits of each upstream database. The project README shows local configuration through Claude's MCP command, but it does not document environment variables or credential settings.

Tools and capabilities

The musharna/plant-genomics-mcp MCP server includes tools for:

• Ensembl Plants and Phytozome locus records, cross-references, sequences, interval queries, and variants.
• UniProt resolution, GO annotations, Plant Ontology, Trait Ontology, and experimental-condition annotations.
• Europe PMC literature, GeneRIF statements, KEGG and PlantCyc pathways, and g:Profiler enrichment.
• Gramene and OrthoDB homology, PANTHER families, STRING interactions, curated experimental interactions, and BAR expression or interaction data.
• AlphaFold and PDBe structures, InterPro domains, JASPAR transcription-factor profiles, and motif matrices.
• NCBI BLAST, VEP variant consequence annotation, and AraGWAS associations.

Coverage varies by backend. Arabidopsis has the broadest support; some services support 12 plants, while others support only Arabidopsis, Arabidopsis and rice, or a smaller organism set. KEGG, AIV, BAR, ThaleMine, GeneRIF, and AraGWAS calls have specific organism or identifier restrictions described by their tools.

Limitations and notes

The musharna/plant-genomics-mcp MCP server depends on live external services, so results can change as those databases update or become unavailable. Returned lists may be capped, with count or truncation fields indicating additional records. A locus identifier may be valid in one namespace but rejected by another; rice AIV, for example, expects MSU LOC_Os* identifiers rather than RAP-DB IDs. KEGG currently resolves only Arabidopsis through the documented cross-backend contract. Unsupported organisms generally produce a typed error, whereas a supported gene with no annotation commonly returns an empty result or found=false.