MCP Server Tutorial

Learn how to use CatGo's MCP (Model Context Protocol) server to integrate with AI assistants like Claude.

Overview

CatGo implements the Model Context Protocol, allowing AI assistants to interact with CatGo's tools for structure analysis, manipulation, and visualization.

Step 1: Start the Server

# MCP is exposed automatically at /api/mcp/ — no flag needed.
# Run the backend (or just launch the CatGo desktop app, which bundles it):
python server/main.py

Step 2: Configure Your AI Client

Add CatGo as an MCP server in your AI client's configuration (e.g., Claude Desktop).

Point the client at the URL (no --mcp, no Python on the client):

http://localhost:8000/api/mcp/

Step 3: Available Tools

The MCP server exposes CatGo's capabilities as tools:

Structure Tools

• Load and parse structure files
• Query structure properties
• Generate slabs and supercells

Analysis Tools

• Compute RDF, band structure, DOS
• Run structure optimization
• Detect symmetry

Workflow Tools

• Create and run computational workflows

Step 4: Example Interactions

Through an MCP-enabled AI assistant:

• "Load the POSCAR file and show the space group"
• "Generate a (111) slab with 3 layers"
• "Set up a VASP relaxation workflow"

Related

• MCP Server Module — Architecture reference
• REST API — Direct HTTP API access