The conversions and property calculations are the daily drivers: hand it a name or a SMILES string and get back the other representations, a formula, or a picture. The two model-backed tools — description in, structure out and the reverse — need extra dependencies and a model key, so a base install quietly leaves them out.
A set of cheminformatics tools over RDKit, PubChem and SELFIES, exposed as MCP tools and selected by name when the server starts.
- Properties from a SMILES string: MoleculeWeight, MoleculeAtomCount, Smiles2Formula and FunctionalGroups
- Name and format conversion in both directions: Iupac2Smiles, Smiles2Iupac, Name2Smiles, Smiles2Cas, Selfies2Smiles and Smiles2Selfies
- SmilesCanonicalization, plus MoleculeSmilesCheck and ReactionSmilesCheck for validating what someone handed you
- MoleculeSimilarity scores two molecules by Tanimoto similarity
- MoleculeVisualizer renders a two-dimensional structure image
- MoleculeModifier proposes chemically modified variants of a molecule
- MoleculeCaptioner and MoleculeGenerator take the language route: a written description of a structure, or a structure from a description
- WebSearch pulls in chemistry information from the web
A Python environment with RDKit and the other cheminformatics libraries installed. The web search tool needs a search-provider key and the two model-backed tools need a language-model key; the rest work without credentials. A ChemSpace key is optional and used only as a fallback when a PubChem lookup misses.
Build from source — clone the repository and build it, then point your client at the binary