Pdbe

What the Pdbe MCP server does

The Pdbe MCP server exposes selected PDBe data to MCP-compatible AI clients. Its tools accept a four-character PDB identifier, such as 1cbs, and return structured information about the corresponding entry. The service is intended for looking up experimental structure metadata, identifying the molecules present in a structure, and connecting PDB chains to UniProt records.

The data source is PDBe, the Protein Data Bank in Europe at EMBL-EBI. PDBe is presented in the documentation as the EBI counterpart to RCSB PDB. All three listed tools are keyless, so the examples do not require an API credential.

How it works

A client sends an MCP tool call with a PDB ID. get_summary returns the entry title, experimental method, release date, source organism, and counts for entities such as polypeptides, ligands, and water. get_molecules lists the entities in the structure, including proteins, nucleic acids, and ligands, with an entity ID, molecule type, name, and chain length.

uniprot_mappings uses PDBe's SIFTS cross-database mappings. Its results associate UniProt accessions with matching PDB chains and residue ranges. This is useful when an agent needs to move from a structural entry to the corresponding UniProt sequence record or identify which part of a chain is covered by a mapping.

The hosted endpoint is served through the Pipeworx gateway. A pack-specific gateway connection lists the PDBe tools along with shared Pipeworx meta-tools. The gateway also offers ask_pipeworx, which can route questions across the wider catalog. The standalone package exposes only the PDBe tools.

Setup and configuration

For a hosted connection, add the following endpoint to an MCP client configuration:

{
  "mcpServers": {
    "pdbe": {
      "url": "https://gateway.pipeworx.io/pdbe/mcp"
    }
  }
}

The README names Claude Desktop, Cursor, and Windsurf as supported configuration targets. A full Pipeworx gateway endpoint is also available at https://gateway.pipeworx.io/mcp when direct access to all catalog packs is preferred.

For local use, run the npm package as a stdio MCP server:

{
  "mcpServers": {
    "pdbe": {
      "command": "npx",
      "args": ["-y", "@pipeworx/mcp-pdbe"]
    }
  }
}

The package can also be started directly with npx -y @pipeworx/mcp-pdbe. The hosted service can be called without an MCP client through the Pipeworx HTTP tools endpoint, using a request such as POST https://gateway.pipeworx.io/v1/tools/pdbe_get_summary with {"pdb_id":"1cbs"}.

Tools and capabilities

• get_summary: retrieves high-level metadata and entity counts for a PDB entry.
• get_molecules: enumerates proteins, nucleic acids, and ligands with entity and chain information.
• uniprot_mappings: returns UniProt accessions, matching chains, and mapped residue ranges for a structure.

Limitations and notes

The documented tools require a four-character PDB ID and are read-only lookup operations. They are keyless according to the tool descriptions. The hosted pack endpoint includes shared gateway meta-tools in addition to the three PDBe tools, so its reported tool count is higher than the standalone package's list. The gateway's exact scope is reported during MCP initialization and may change.