The two entry tools are what make this useful in a conversation: you rarely know a Reactome stable id, you know a gene symbol or a disease name. From there the hierarchy, participants, reactions and interaction tools let the model descend without you translating identifiers by hand. It is unofficial — a community wrapper over Reactome's public service, not a Reactome product.
An unofficial MCP server over the Reactome Content Service, the curated pathway and systems-biology database. It answers the questions a researcher actually starts from — which pathways contain this gene, which pathways relate to this disease — and then lets you descend into a pathway's molecules, reactions and interaction network by its stable identifier.
- Pathways found by name, description or keyword, optionally narrowed to pathway, reaction, protein, complex or disease — `search_pathways`
- One pathway's full record by its Reactome stable identifier — `get_pathway_details`
- Every pathway containing a gene symbol or UniProt ID, with species selectable and human as the default — `find_pathways_by_gene`
- Disease-associated pathways by disease name or DOID identifier — `find_pathways_by_disease`
- A pathway's parent and child structure, so you can see where it sits — `get_pathway_hierarchy`
- All molecules participating in a pathway, and all biochemical reactions within it — `get_pathway_participants`, `get_pathway_reactions`
- Protein-protein interactions within a pathway, filterable to protein-protein, regulatory, catalysis or all — `get_protein_interactions`
- Resource templates for direct addressing: `reactome://pathway/{id}`, `reactome://reaction/{id}`, `reactome://protein/{id}`, `reactome://disease/{id}`, `reactome://search/{query}`
No account and no key — the Reactome Content Service at `https://reactome.org/ContentService` requires no authentication and its limits are described as generous for research use. Build it yourself: `npm install`, `npm run build`, then point your client at `build/index.js`. Calls carry a 30s timeout.