AI assistants are becoming a common way to look for scientific information. They are only as reliable as the data behind their answers, though. A language model relying on memory can provide outdated, incomplete, or fabricated details about a structure, ligand, or annotation. The Model Context Protocol (MCP) is an open standard for connecting AI agents directly to authoritative data sources. Through MCP, an agent can query the archive itself and answer with current, traceable information behind PDB entries.
We are releasing the beta version of the RCSB PDB MCP server. It gives AI agents access to three RCSB PDB APIs and additional helper tools:
- Search API: find structures by keyword, attributes, sequence similarity, chemical structure, 3D shape, and sequence or structural motifs, combined with Boolean logic and summarised with counts and facets
- Data API: retrieve detailed information on entries, entities, assemblies, chains, ligands and interfaces, plus linked UniProt and PubMed records
- Sequence Coordinates API: map sequences and positional features between PDB, UniProt and NCBI
- Helper tools turn free-text terms into the identifiers the archive is indexed by: Gene Ontology, InterPro/Pfam, EC numbers, MONDO disease terms, and NCBI Taxonomy
The server works with any MCP-compatible client, such as Claude, Cursor, or ChatGPT, and it is open source.
For example, you can ask an agent Is there any protein interacting with uranium compounds?
followed by Can you list the residue binding sites for the results?
and What are the most frequent amino acid types in these binding sites?
This beta release is still under active development. Tools and behavior may change, and key results should be checked on RCSB.org.
We would very much like your help, particularly in letting us know
- questions the agent answered well or got wrong
- the client and model used
- Suggestions for additional data or features
Your feedback will guide how we improve the current tools and what we build next.
Connect: https://mcp-beta.rcsb.org/mcp
Source: https://github.com/rcsb/rcsb-mcp
Feedback: https://github.com/rcsb/rcsb-mcp/issues (or info@rcsb.org)
Many thanks to our summer scholars Hriday Adani and Yusuf Mohammed for their help on this project as part of the Chancellor Signature Initiatives Summer Undergraduate Research Fellowship at the Rutgers Artificial Intelligence and Data Science (RAD) Collaboratory.