The value is that SMILES, InChI Key, CAS and ATC come back as fields — a model can pass them to the next tool instead of re-deriving them, and the PubChem link means the answer is checkable. Scope is exactly that: identity and basic properties, with no pharmacology, interactions or clinical data.
A small MCP server that queries the PubChem API for the basic chemical facts about a drug. You give it a name; it returns the identifiers and properties in a structured object, with a link to the PubChem compound page so the answer can be checked.
- Molecular formula and molecular weight
- Structure identifiers a downstream tool can consume: SMILES and the InChI Key
- Registry and classification codes: the CAS Number and the ATC Code
- The IUPAC name and a synonym list, which is what resolves the same compound appearing under several trade names
- A `Details Link` to the compound's PubChem page, so every answer is verifiable at the source
No account and no key — PubChem's API is public. Python 3.10, then either `pip install .` from a clone or run the published package directly with `uvx pubchem_mcp_server`. The server configuration follows the same shape as Claude Desktop's, so it slots into an existing `servers_config.json` alongside other servers.
One command — uvx pubchem_mcp_server
