PDB — 3D protein structure search via MCP

Connect

MCP endpoint

https://mcp.polarisray.com/pdb

Transport: http · Auth: none · See the connection guide for client-specific steps.

Tools

search_structures Search PDB entries by free text and filters — source organism, experimental method (X-ray / electron microscopy / NMR), polymer type, bound ligand (chemical component id) or ligand InChIKey, UniProt accession, taxonomy id, and resolution or release-date ranges.
semantic_search_structures Search structures by meaning, not just exact keywords — finds relevant entries by concept across title, organism and ligand context, blended with keyword search.
get_structure Retrieve the full record for a single entry by its PDB ID — title, method, resolution, polymer entities (sequence, organism, UniProt), bound ligands (id, name, InChIKey, SMILES), plus the RCSB entry URL and mmCIF / BinaryCIF coordinate download links.
aggregate_structures Count and group structures by experimental method, organism, polymer type, bound ligand, or UniProt target — e.g. the most-represented ligands in human structures, or a release-date trend.
resolve Normalize lay terms, synonyms and abbreviations to standard concepts (e.g. "spike protein", "kinase") for precise filtering across PolarisRay's sources.

Example questions

"Find high-resolution human kinase structures bound to an inhibitor."
"What structures contain the heme ligand, and from which organisms?"
"Show cryo-EM structures of the SARS-CoV-2 spike protein."
"Look up PDB entry 4HHB and list its chains, organism and bound ligands."
"Which UniProt targets have the most experimental structures?"

About this dataset

The Protein Data Bank (PDB) is the single global archive of experimentally-determined three-dimensional structures of proteins, nucleic acids and their complexes — the structural foundation of molecular biology and structure-based drug design, maintained by the Worldwide PDB (wwPDB) and served here via RCSB. This connector indexes one record per entry (keyed by its stable PDB ID, e.g. 4HHB), rolling up each entry’s experimental method and resolution, its polymer entities (one-letter sequences, source organism and taxonomy, UniProt accessions) and its bound ligands (chemical component id, name, InChIKey, SMILES), and is refreshed from the weekly wwPDB release.

Where ChEMBL and PubChem describe the chemistry of compounds, PDB supplies the 3D structural context — which proteins a ligand binds, from which organism, at what resolution. A structure’s bound-ligand InChIKey is the join key back to those chemistry sources, and its UniProt accession is the join to drug targets, so an assistant can pivot from a compound to the structures it appears in, or from a target to every structure solved for it.

What you can ask

Ask plain-English questions and the assistant translates them into structured searches: filter by source organism, experimental method (X-ray diffraction, electron microscopy, solution NMR), polymer type (protein / DNA / RNA), bound ligand (by chemical component id such as HEM, or by InChIKey), or UniProt accession; constrain by resolution or release date; retrieve a full entry by its PDB ID; or aggregate across the archive to see, for example, which UniProt targets have the most solved structures. Every hit carries the official RCSB entry url plus direct mmCIF and BinaryCIF coordinate download links.

Search by meaning

In addition to keyword and structured search, you can search structures by meaning — over the entry title, organism and ligand context. Ask for “structures of the coronavirus spike protein” or “antibody–antigen complexes” and it surfaces conceptually related entries even when the title wording differs, with the same structured filters (organism, method, resolution, …) still applying.

Caveats

This connector indexes structure metadata, not the atomic coordinates: it links to the mmCIF / BinaryCIF files on RCSB rather than redistributing them. Coverage is experimental structures only — computed models (AlphaFold, ModelArchive) are out of scope. Resolution and method vary widely across the archive, and older entries may lack fields such as R-free; counts are counts of indexed entries, not of atoms or depositions. PDB data are CC0 public domain, but citing the original depositors is good practice. Always confirm details against the primary RCSB entry, and never treat structural data as medical advice.