Pubchem MCP Server

What pubchem-mcp-server does

The pubchem-mcp-server MCP server connects MCP-compatible clients to PubChem’s compound and bioassay data. It lets an agent identify compounds from names, SMILES, InChIKeys, formulas, substructures, superstructures, or 2D similarity searches. Searches can also return compounds marked active by a specified bioassay.

Once a compound CID is known, the server can return physicochemical properties, descriptions, synonyms, classifications, drug-likeness checks, safety information, images, 3D conformers, interactions, bioactivity, and external database references. It also supports searches for assays by gene symbol, protein, Gene ID, or UniProt accession, plus summaries for assays, genes, proteins, and taxonomy entities.

How it works

The server exposes 10 MCP tools and six URI-templated resources. Compound resources include core properties, GHS safety classifications, 2D images, cross-references, and bioactivity. An assay resource returns an assay summary. Clients that only expose tools may not show these resources.

Search and retrieval operations use PubChem identifiers and paging controls. Compound searches can return up to 200 CIDs per page, while detail lookups accept up to 100 CIDs per call and safety lookups accept up to 25. The responses report unresolved identifiers, capped inputs, truncation, totals, and per-record status where relevant. This allows an agent to distinguish missing records, incomplete pages, absent classifications, and compounds with no available data.

The 3D structure tool returns parsed atoms and bonds as JSON or the raw V2000 SDF text. JSON previews can cap atoms and bonds while still reporting full counts. A compound may lack a computed 3D structure, including some large molecules, mixtures, and salts.

Setup and configuration

The pubchem-mcp-server MCP server can run through the published npm package with npx -y @cyanheads/pubchem-mcp-server, as shown by the repository’s client installation links. The project supports STDIO and Streamable HTTP. A public hosted endpoint is available at https://pubchem.caseyjhand.com/mcp.

Named installation paths are provided for Claude Desktop, Cursor, and VS Code. Other MCP clients can connect when configured for the server’s supported transport.

Tools and capabilities

• Search compounds by identifiers, formulas, structure relationships, similarity, or assay activity.
• Retrieve properties, descriptions, synonyms, FDA/MeSH/ATC classifications, and Lipinski/Veber checks.
• Fetch 2D PNG diagrams and 3D conformers in JSON or SDF form.
• Retrieve GHS signal words, pictograms, hazard codes, precautionary codes, and source classifications.
• Inspect bioassay results and filter them by outcome or molecular target.
• Find drug-drug, drug-food, and chemical-target interactions.
• Retrieve references to registry IDs, CAS numbers, patents, PubMed, genes, proteins, and taxonomy records.

The pubchem-mcp-server MCP server also exposes summaries for assays, genes, proteins, and taxonomy entities. Bioactivity filters can use a target Gene ID or accession, and matching identifiers associated with the same target can be reached through that filter.

Limitations and notes

PubChem-backed results may be paged, capped, truncated, or unavailable for a particular entity. Description and classification enrichment in compound details covers only the first 10 CIDs, while synonyms and drug-likeness checks cover every found CID. Images and 3D structures operate on one CID at a time.

Safety responses distinguish a nonexistent CID from a compound without deposited GHS data or one for which no notifier reported a GHS hazard. An empty cross-reference response can mean either that the requested reference types are absent or that the CID may be incorrect. The repository notes that many MCP clients expose tools but not resources.