Resolve chemical names to PubChem identifiers and retrieve traceable compound properties from Claude Desktop through a locally installed PubChem MCP server.
Level: IntermediateCost: MixedPrivacy: Cloud~30 min
Retrieve compound details while preserving sources and missing values.
What you'll build
Scope
Use @cyanheads/pubchem-mcp-server 0.6.5 with Node.js >=24. The MCP process runs locally; it queries public PubChem PUG REST/PUG View APIs, and the conversation runs through Claude. No PubChem API key is required for this baseline. This is compound-information retrieval, not an experimentally validated property predictor. Record the server version and the retrieval date. PubChem XLogP must not be relabelled as RDKit Crippen logP.
Install Node.js 24 or newer with npm and Claude Desktop. Confirm that local MCP is allowed and HTTPS access to the target database is available. Bun is not needed for this Node-based variant.
Use a new writable working directory. This installs the server in .mcp-packages, rather than globally; keep its lockfile for verification. The upstream npx entry and the local dist/index.js entry use the same published package.
Open Settings > Developer > Edit Config and merge the server into mcpServers. Replace all placeholder paths with actual absolute paths, including the node executable. Keep existing server entries. No API keys are needed for the baseline query; optional keys/emails belong in private configuration.
Valid JSON starts the installed package with the Node 24+ executable and stdio.
4. Configure Claude Desktop on Windows
Windows
Open Settings > Developer > Edit Config and merge the server into mcpServers. Replace all placeholder paths with actual absolute paths, including the node executable. Keep existing server entries. No API keys are needed for the baseline query; optional keys/emails belong in private configuration.
Valid JSON starts the installed package with the Node 24+ executable and stdio.
5. Restart and inspect the tools
All platforms
Completely quit and restart Claude Desktop. Inspect Developer status and the connector tool list. Required tools: pubchem_search_compounds, pubchem_get_compound_details.
Connection succeeds and the listed tools are available.
6. Resolve and inspect aspirin
All platforms
Paste into a new chat and inspect both tool calls.
Use pubchem_search_compounds to resolve the name aspirin, then pubchem_get_compound_details for the returned CID. Report the CID, formula, molecular weight, TPSA and XLogP when returned, the original PubChem link and retrieval date. Check whether aspirin resolves to CID 2244. Separate computed properties from experimental evidence; preserve missing fields. Do not infer toxicity or treatment recommendations.
Expected baseline: aspirin CID 2244, formula C9H8O4, weight about 180.16 g/mol. Other fields must come from returned data, with units and source URL https://pubchem.ncbi.nlm.nih.gov/compound/2244 . These are expected checks, not session measurements.
7. Check a missing identifier
All platforms
Repeat with an intentionally nonexistent name.
Search the compound name ChemAIAtlas-no-such-compound-20261007. Report no match or the actual API error; do not fabricate a CID.
Use the PubChem website or PUG REST directly for the same CID; local RDKit calculations are a separate alternative and use different descriptor definitions.