Uniprot MCP Server

What uniprot-mcp-server does

The uniprot-mcp-server MCP server gives an MCP-compatible AI client access to UniProtKB protein and organism data. It supports searching by plain text or Lucene field queries, retrieving curated entries, translating identifiers from several external databases, and reading sequences. Additional tools cover reference proteomes and NCBI taxonomy records.

Search results foreground reviewed Swiss-Prot records by default and can include facet counts. Entry retrieval exposes structured sections such as function, catalytic activity, disease, variants, isoforms, Gene Ontology terms, and cross-references. The server is suited to the Biology & Bioinformatics category when an agent needs protein annotations, accession discovery, or organism context.

How it works

The uniprot-mcp-server MCP server communicates with UniProtKB through MCP tools and supports either STDIO or Streamable HTTP transport. Search and proteome protein lists use forward cursor pagination. Cursor validity depends on the relevant query and tool, so a changed query or invalid token requires starting a new page sequence.

Identifier conversion uses UniProt’s asynchronous ID-mapping service. A request may return a running job ticket; completed jobs return result pages and may provide a continuation value for the next page. Gene-symbol mappings can be paired with an NCBI taxon ID to select the intended species.

Two resources mirror individual records: uniprot://entry/{accession} and uniprot://taxonomy/{taxonId}. A uniprot_protein_dossier prompt describes a guided workflow for resolving an identifier, obtaining a curated entry, and reviewing disease, variant, structural, citation, and bioactivity references.

Setup and configuration

The package is published as @cyanheads/uniprot-mcp-server. The repository provides installation paths for Claude Desktop, Cursor, and VS Code, including an npx -y @cyanheads/uniprot-mcp-server configuration for clients that launch STDIO servers. A hosted MCP endpoint is also listed at https://uniprot.caseyjhand.com/mcp.

Client configuration should select either the local package-based transport or the hosted Streamable HTTP endpoint, depending on the client’s MCP support.

Tools and capabilities

The six tools are:

• uniprot_search_proteins: discovers proteins with text or Lucene queries, optional reviewed filtering, facets, and cursor pagination.
• uniprot_get_entry: retrieves up to 20 accessions per call with partial-success reporting and selectable sections.
• uniprot_map_ids: maps gene names, Ensembl, RefSeq, ChEMBL, PDB, GeneID, and related identifiers to UniProtKB targets.
• uniprot_get_proteome: returns proteome metadata and, optionally, a capped filtered protein page.
• uniprot_get_taxonomy: resolves a taxon by NCBI taxon ID or scientific name and can return immediate children.
• uniprot_get_sequence: returns canonical FASTA data, sequence length, parsed headers, and optional isoforms.

Large entry responses return an outline so the client can request specific sections. Inactive accessions and unmapped identifiers are reported with status or replacement information rather than being treated as ordinary successful results.

Limitations and notes

The uniprot-mcp-server MCP server does not enumerate the full UniProtKB corpus through resources; search is the discovery mechanism. Proteome protein data is opt-in and capped, and large records may require section-specific follow-up calls. uniprot_get_proteome requires exactly one of a UPID or NCBI taxon ID. Identifier-mapping jobs and paginated responses can expire or become invalid, requiring a restart. Sequence retrieval is accession-based, and entry requests should remove isoform suffixes before fetching the parent record.