Gprofiler MCP server

@pipeworx/gprofiler

Functional enrichment for a gene list against GO, KEGG, Reactome, WikiPathways, TRANSFAC, miRTarBase, CORUM, HPA and HPO, plus gene/protein identifier conversion and cross-species ortholog mapping — from g:Profiler at the University of Tartu.

Part of Pipeworx — an MCP gateway connecting AI agents to 1679+ live data sources.

Tools

— g:GOSt. Returns each enriched term with an ALREADY-ADJUSTED p-value (g:SCS), term size, query size, overlap, precision and recall. Per-gene intersections and ontology parents are dropped: they dominate the payload and nothing reads them in an answer.

• gprofilerenrich(organism, query[], sources?, userthreshold?, noiea?, ordered?, domainscope?, background?, significant?, limit?)

gene/protein/transcript/probe namespaces and reports hitsForInput, so an ambiguous symbol shows up as ambiguous instead of quietly resolving to one gene.

• gprofilerconvertids(organism, query[], target?, numeric_namespace?) — g:Convert. Maps between

mouse Trp53, with one-to-many mappings flagged rather than collapsed.

• gprofiler_orthologs(organism, target, query[]) — g:Orth via Ensembl Compara. Human TP53 →

Auth

Keyless. g:Profiler asks programmatic callers to identify themselves; the pack sends a pipeworx-mcp-gprofiler User-Agent.

Data sources

• POST https://biit.cs.ut.ee/gprofiler/api/gost/profile/ — enrichment.
• POST https://biit.cs.ut.ee/gprofiler/api/convert/convert/ — ID conversion.
• POST https://biit.cs.ut.ee/gprofiler/api/orth/orth/ — orthologs.
• Docs:

Traps

The organism code is g:Profiler's own and nothing else works. First letter of the genus plus the full species name, lowercase: hsapiens, mmusculus, rnorvegicus, drerio, dmelanogaster, celegans, scerevisiae, athaliana. "human" and "9606" are both rejected. This is the common first failure, and it is a loud one, which is the good case.

A single gene passed as a bare string would be split into characters upstream. The pack accepts a string and splits it on whitespace/comma/semicolon before sending, so "TP53" becomes ["TP53"] rather than four failed lookups returned as a clean 200.

An empty enrichment result is ambiguous and the pack says so. No significant term is a real answer for a small or functionally unrelated list — and it is also exactly what a wrong organism code or unrecognised identifiers produce. The note field points the caller at gprofilerconvertids to tell the two apart.

pValue is already multiple-testing corrected (g:SCS by default). Do not correct it again.

nincoming > 1 on a conversion means the INPUT matched more than one record**, not that the output is multi-valued. Surfaced as hitsForInput and collected in ambiguousInputs.

Enrichment against all sources on a large list is genuinely slow (seconds, not milliseconds). Pass sources when you know which annotation set you want.

Quick Start

Add to your MCP client (Claude Desktop, Cursor, Windsurf, etc.):

{
  "mcpServers": {
    "gprofiler": {
      "url": "https://gateway.pipeworx.io/gprofiler/mcp"
    }
  }
}

What this endpoint actually serves

tools/list at https://gateway.pipeworx.io/gprofiler/mcp returns the tools in the table above plus the shared Pipeworx meta-tools — askpipeworx, discovertools, search_within, remember/recall and the rest of the gateway-wide set. So the tool count you see is larger than this table: a single-pack endpoint currently lists roughly 30 shared tools alongside the pack's own. The connection's initialize response states its exact scope, and is the authoritative answer for a given day.

Shortened. The full README is on GitHub.