The scope is deliberately narrow: look up an active site by PDB ID, or find structures tied to a disease. That is enough to stop a model inventing residue numbers, and it is the piece most biomedical chats are missing. Anything beyond those two questions — literature, structure prediction, 3-D coordinates — is roadmap, not shipped.
An MCP server that puts protein structure knowledge from the RCSB Protein Data Bank in front of a model, so it can answer structural biology questions without that data being in its training set. The server handles the API calls, formatting and error handling.
- `analyze-active-site` — binding sites and functional residues for a protein, given a PDB ID
- `search-disease-proteins` — protein structures associated with a named disease or medical condition
Node.js 20.0.0 or higher, TypeScript 5.0+, and an MCP client. There is no published package: clone the repo, `npm install`, `npm run build`, then point your client at `node` and the built `build/index.js`. Smithery can install it for Claude Desktop in one command. No account or API key.
One command — npx -y @smithery/cli install @acashmoney/bio-mcp --client claude