PDBe MCP Servers

A set of Model Context Protocol (MCP) servers that provides seamless access to the Protein Data Bank in Europe (PDBe) API and PDBe Search. These servers expose PDBe's comprehensive structural biology data as MCP tools, enabling direct integration with any AI client that supports MCP.

The package also includes an advanced PDBe Graph server for users who run their own local PDBe-KB Neo4j graph database. PDBe does not provide a public running graph database instance for this MCP server to query, so most users should start with the API and Search servers.

Features:

• PDBe API Server: Access core structural data through REST API endpoints

• PDBe Search Server: Perform advanced Solr-based searches across structural data

• PDBe Graph Server: Inspect the graph schema and, with a local PDBe-KB Neo4j setup, query complex relationships and molecular interactions

Prerequisites

• Python 3.10+ - Required for the server runtime

• uv - Fast Python package manager and dependency resolver

Related MCP server: UniProt MCP Server

Installation

Quick Start

Run directly from PyPI:

uvx pdbe-mcp-server

The tool is available on PyPI and can be run directly with uvx without any installation step.

Alternative: Local Development Installation

For development work or customization:

1. Clone and navigate to the repository:

   git clone https://github.com/PDBeurope/PDBe-MCP-Servers.git
   cd PDBe-MCP-Servers

2. Create a virtual environment:

   uv venv

3. Install with uv:

   uv pip install .

AI Client Integration

Configuration

1. Open your AI client's MCP configuration.

   MCP-compatible clients use different settings locations and file formats. Many JSON-based clients use an mcpServers object, while some clients provide commands or a settings UI for adding servers.

2. Add the recommended PDBe MCP server configuration.

   For JSON-based clients that support mcpServers, add:

   For PyPI installation (recommended):

   {
     "mcpServers": {
       "PDBe API Server": {
         "command": "uvx",
         "args": [
           "pdbe-mcp-server",
           "--server-type",
           "pdbe_api_server"
         ]
       },
       "PDBe Search Server": {
         "command": "uvx",
         "args": [
           "pdbe-mcp-server",
           "--server-type",
           "pdbe_search_server"
         ]
       }
     }
   }

   For local development installation:

   {
     "mcpServers": {
       "PDBe API": {
         "command": "/usr/local/bin/uv",
         "args": [
           "run",
           "--directory",
           "/path/to/your/PDBe-MCP-Servers",
           "pdbe-mcp-server",
           "--server-type",
           "pdbe_api_server"
         ]
       },
       "PDBe Search": {
         "command": "/usr/local/bin/uv",
         "args": [
           "run",
           "--directory",
           "/path/to/your/PDBe-MCP-Servers",
           "pdbe-mcp-server",
           "--server-type",
           "pdbe_search_server"
         ]
       }
     }
   }

Note:

• For the PyPI installation method, ensure uvx is available in your PATH (this comes with uv)

• For local development, ensure that uv is installed and the /path/to/your/PDBe-MCP-Servers matches your actual directory

Add the graph server only if you have a local PDBe-KB Neo4j graph database configured. See Advanced Graph Server Configuration.

3. Restart or reload your AI client to load the new configuration.

Antigravity Example

In Antigravity, open Manage MCP Servers and select View raw config, or edit ~/.gemini/antigravity/mcp_config.json, then add the PDBe server entries:

{
  "mcpServers": {
    "PDBe API Server": {
      "command": "uvx",
      "args": [
        "pdbe-mcp-server",
        "--server-type",
        "pdbe_api_server"
      ]
    },
    "PDBe Search Server": {
      "command": "uvx",
      "args": [
        "pdbe-mcp-server",
        "--server-type",
        "pdbe_search_server"
      ]
    }
  }
}

Codex Example

In Codex, add the PDBe MCP servers with the CLI:

codex mcp add pdbe-api -- uvx pdbe-mcp-server --server-type pdbe_api_server
codex mcp add pdbe-search -- uvx pdbe-mcp-server --server-type pdbe_search_server
codex mcp list

For a local development checkout, point Codex at the repository directory:

codex mcp add pdbe-api-local -- uv run --directory /path/to/your/PDBe-MCP-Servers pdbe-mcp-server --server-type pdbe_api_server
codex mcp add pdbe-search-local -- uv run --directory /path/to/your/PDBe-MCP-Servers pdbe-mcp-server --server-type pdbe_search_server

Using in an AI Client

Once configured, you can access PDBe tools directly in your AI client conversations:

• Search for protein structures: "Find structures for UniProt accession P12345"

• Query structure releases: "Show me all structures released this month grouped by experimental method"

• Advanced search queries: "Find all X-ray crystal structures with resolution better than 2.0 Å from 2024"

The tools will appear in your AI client's tools interface, where you can enable or disable them as needed.

Server Types

• pdbe_api_server: Core PDBe REST API access with essential structural data

• pdbe_search_server: Advanced Solr-based search capabilities for complex structural queries and data analysis

• pdbe_graph_server: Advanced/local server for inspecting the PDBe-KB graph schema and optionally executing read-only Cypher queries against a locally configured Neo4j database

Tool Reference

API Server Tools

The pdbe_api_server generates tools from the PDBe API OpenAPI specification. Use this server for core PDBe REST API data, such as entries, assemblies, molecules, ligands, publications, and validation information.

Search Server Tools

get_search_schema

Retrieves the complete Solr search schema showing all available fields, data types, and descriptions. Use this to understand what fields you can search and filter on.

Example usage:

"Show me the search schema for PDBe structures"

run_search_query

Execute Solr-style search queries with flexible field selection, filter queries, facets, grouping, sorting, and pagination options.

Parameters:

• query (required): Raw Solr query string passed as q (e.g., *:*, pdb_id:1cbs, text:*kinase*, resolution:[0 TO 2.0])

• fl (optional): Field list as a string or array of field names to include in results

• filters (optional): Backwards-compatible alias for fl

• fq (optional): Filter query string or array of filter query strings

• sort (optional): Sort criteria (e.g., release_date desc, resolution asc)

• start (optional): Starting index for pagination (default: 0)

• rows (optional): Number of results to return (default: 10)

• facet (optional): Enable Solr faceting

• facet_fields (optional): Field facet string or array, sent as facet.field

• facet_queries (optional): Query facet string or array, sent as facet.query

• facet_limit, facet_mincount, facet_sort (optional): Common facet controls

• group (optional): Enable Solr grouping

• group_field (optional): Grouping field string or array, sent as group.field

• group_limit, group_offset, group_sort (optional): Common grouping controls

• params (optional): Object of additional Solr parameters for advanced use

Example queries:

{
  "query": "*:*",
  "fq": ["release_date:[2025-10-01T00:00:00Z TO 2025-10-31T23:59:59Z]"],
  "group": true,
  "group_field": "experimental_method",
  "rows": 0
}

{
  "query": "*:*",
  "fq": ["experimental_method:\"X-ray diffraction\"", "resolution:[0 TO 2.0]"],
  "fl": ["pdb_id", "title", "resolution", "experimental_method"],
  "sort": "resolution asc",
  "rows": 20
}

{
  "query": "text:*ATP*",
  "facet": true,
  "facet_fields": ["ligand_name", "experimental_method"],
  "facet_mincount": 1,
  "rows": 10
}

Search Field Examples

Common searchable fields include:

• pdb_id: PDB entry identifier

• experimental_method: Structure determination method

• release_date: Structure release date

• resolution: Structure resolution (Å)

• molecule_type: Type of molecule (protein, DNA, RNA, etc.)

• organism_scientific_name: Source organism

• ligand_name: Bound ligands

• title: Structure title/description

Use get_search_schema to discover all available fields and their descriptions.

Development and Advanced Usage

Development Installation

For contributing or development work, first clone the repository and then install in editable mode:

git clone https://github.com/PDBeurope/PDBe-MCP-Servers.git
cd PDBe-MCP-Servers
uv sync --all-extras --dev

Node.js (optional) - For using the MCP Inspector development tool

Starting the Server Manually

Most users should run the API server, the Search server, or both.

PDBe API Server

Provides access to core PDBe REST API endpoints:

Using PyPI installation:

uvx pdbe-mcp-server --server-type pdbe_api_server --transport sse

Using local development:

uv run pdbe-mcp-server --server-type pdbe_api_server --transport sse

PDBe Search Server

Provides advanced Solr-based search and analytics capabilities:

Using PyPI installation:

uvx pdbe-mcp-server --server-type pdbe_search_server --transport sse

Using local development:

uv run pdbe-mcp-server --server-type pdbe_search_server --transport sse

The server will start at http://localhost:8000/sse by default.

Advanced Graph Server Configuration

The pdbe_graph_server is intended for users who have downloaded and configured the PDBe-KB graph database in their own environment. PDBe does not provide a public running Neo4j instance for this MCP server to query.

To set up the graph database locally, follow the PDBe-KB graph documentation: https://www.ebi.ac.uk/pdbe/pdbe-kb/graph

Once your local Neo4j database is running, set these environment variables before starting the graph server:

• NEO4J_URL: The Neo4j database URL (e.g., bolt://localhost:7687)

• NEO4J_USERNAME: The Neo4j username

• NEO4J_PASSWORD: The Neo4j password

• NEO4J_DATABASE (optional): The database name. When set, this is passed to the Neo4j driver for Neo4j 4+. For Neo4j 3.5 compatibility, omit this variable to use the default database.

The Neo4j driver is included in this package's dependencies.

MCP Client Graph Configuration

Add this server only when the environment variables above are available to your AI client.

For PyPI installation:

{
  "mcpServers": {
    "PDBe Graph Server": {
      "command": "uvx",
      "args": [
        "pdbe-mcp-server",
        "--server-type",
        "pdbe_graph_server"
      ],
      "env": {
        "NEO4J_URL": "bolt://localhost:7687",
        "NEO4J_USERNAME": "neo4j",
        "NEO4J_PASSWORD": "your-password"
      }
    }
  }
}

For local development installation:

{
  "mcpServers": {
    "PDBe Graph": {
      "command": "/usr/local/bin/uv",
      "args": [
        "run",
        "--directory",
        "/path/to/your/PDBe-MCP-Servers",
        "pdbe-mcp-server",
        "--server-type",
        "pdbe_graph_server"
      ],
      "env": {
        "NEO4J_URL": "bolt://localhost:7687",
        "NEO4J_USERNAME": "neo4j",
        "NEO4J_PASSWORD": "your-password"
      }
    }
  }
}

Codex example:

codex mcp add pdbe-graph \
  --env NEO4J_URL=bolt://localhost:7687 \
  --env NEO4J_USERNAME=neo4j \
  --env NEO4J_PASSWORD=your-password \
  -- uvx pdbe-mcp-server --server-type pdbe_graph_server

Starting the Graph Server Manually

Using PyPI installation:

uvx pdbe-mcp-server --server-type pdbe_graph_server --transport sse

Using local development:

uv run pdbe-mcp-server --server-type pdbe_graph_server --transport sse

Graph Server Tools

pdbe_graph_nodes

Retrieves metadata about all node types (labels) defined in the PDBe graph database schema. This uses the public graph schema and does not require local Neo4j credentials.

Example usage:

"Show me all node types in the PDBe graph database"

pdbe_graph_edges

Retrieves metadata about all relationship types (edges) defined in the PDBe graph database schema. This uses the public graph schema and does not require local Neo4j credentials.

Example usage:

"Show me all relationship types in the PDBe graph database"

pdbe_graph_node_relationships

Verifies selected node labels and returns the incoming, outgoing, and self-loop relationship patterns defined for each label. This uses the public graph schema and does not require local Neo4j credentials.

Parameters:

• node_labels (required): List of exact, case-sensitive node labels to verify.

Example usage:

"Verify relationships for Entry, Entity, and UniProt"

pdbe_graph_example_queries

Retrieves example Cypher queries that demonstrate how to interact with the PDBe graph database. This uses the public graph schema and does not require local Neo4j credentials.

Example usage:

"Give me example Cypher queries for exploring the PDBe graph"

pdbe_run_cypher_query

Execute custom read-only Cypher queries against your configured Neo4j graph database. This tool is only available when Neo4j environment variables are configured.

Parameters:

• cypher_query (required): The Cypher query to execute. Only MATCH and OPTIONAL MATCH queries are allowed.

Example usage:

"Execute query: MATCH (s:Structure) WHERE s.PDB_ID = '1abc' RETURN s.TITLE as title"
"Find ligands: MATCH (s:Structure)-[:HAS_LIGAND]->(l:Ligand) WHERE s.PDB_ID = '1abc' RETURN l.name"

Security: Only read-only queries are allowed (MATCH, OPTIONAL MATCH). Write operations (MERGE, CREATE, DELETE, REMOVE, SET, LOAD CSV, FOREACH) are blocked to prevent accidental data modification.

The tool response is formatted as JSON by default, but can be converted to TOON format by setting TOON_ENABLED=true.

Development and Testing

Explore available tools and test API responses:

npx @modelcontextprotocol/inspector

The MCP Inspector provides an interactive interface to browse tools, test queries, and validate responses before integrating with your application.

Server Configuration

Transport Options

• stdio: Default mode - Optimal for direct MCP client integration

• SSE (Server-Sent Events): --transport sse - Best for web-based clients and development

Experimental TOON Output

You can enable experimental TOON-formatted output for PDBe API tool responses and Neo4j Cypher query results by setting the environment variable TOON_ENABLED=true. See the TOON format specification at https://toonformat.dev/.

• If TOON encoding fails for any reason, the server falls back to JSON output.

• This feature is experimental and intended for opt-in usage only.

Troubleshooting

Common Issues

"Command not found" errors:

• Ensure uv is installed and in your PATH

• Verify the full path to uv in your AI client's MCP configuration

Missing tools in your AI client:

• Restart or reload your AI client after configuration changes

• Check your AI client's MCP server logs for errors

• Verify JSON syntax in your configuration file

Resources

• Model Context Protocol - Official MCP documentation and specifications

• PDBe API Documentation - Complete API reference and examples

• PDBe Graph Database - Advanced querying and relationship mapping

• Antigravity MCP Documentation - MCP setup instructions for Antigravity

• OpenAI Docs MCP - Codex MCP configuration examples

License

This project is licensed under the Apache License, Version 2.0 - see the LICENSE file for details.

Support

For questions, bug reports, or feature requests:

• Issues: Use the GitHub Issues tracker

• PDBe Helpdesk: Visit the PDBe Help & Support pages