CST-MCP: AI Agent Interface for CST Studio Suite

Overview

CST-MCP is an implementation of the Model Context Protocol (MCP) designed to bridge Large Language Models (LLMs) with CST Studio Suite. It empowers AI agents to autonomously perform electromagnetic simulation tasks, including geometric modeling, solver configuration, and results analysis.

This project aims to transform CST Studio Suite into a controllable environment for AI, moving beyond simple scripting to full "agentic" capabilities where the AI can close the loop between design, simulation, and verification.

Project Goal

To build a robust set of tools and an MCP server that allows an AI Agent to:

1. Model: Create and manipulate 3D structures using boolean operations and parametric modeling.
2. Simulate: Configure boundaries, ports, and solvers, and execute simulations.
3. Analyze: Export and interpret results (S-parameters, field monitors) to make design decisions.
4. Iterate: Automatically correct errors (e.g., geometry intersection, mesh failures) and optimize designs.

Features (Planned & Implemented)

• MCP Server: A standardized interface for agents to discover and call CST tools.
• Geometric Primitives: Create bricks, cylinders, spheres, etc.
• Boolean Operations: Union, subtract, intersect for complex modeling.
• Simulation Control: Open projects, set parameters, run solvers (T-Solver, F-Solver).
• Results Export: Extract Touchstone files (S-parameters) and key metrics (bandwidth, resonance).

Structure

• server/: Implementation of the MCP server.
• tools/: Core logic for CST interactions (wrappers around CST Python libraries).
• utils/: Helper functions for geometry and data handling.
• demos/: Example scripts demonstrating automated workflows.

Roadmap

Phase 1: The Automation Loop

• Establish minimal Python automation (Open -> Model -> Solve -> Export).
• Stabilize core tools using official CST Python Libraries or COM interface.

Phase 2: Minimal MCP Server

• Implement open_project, set_parameters, run_solver, export_results tools.
• specific geometry tools (make_brick, boolean_subtract, etc.) for detailed modeling.

Phase 3: Intelligent Feedback

• Experiment Tracking: Log every tool call with complete context for reproducibility.
• Auto-Diagnosis: Detect and classify common failures (mesh errors, port overlaps).
• Semantic Results: Return structured engineering metrics (e.g., "10dB Bandwidth") instead of raw files.

Getting Started

(Instructions on how to set up the environment and run the server will be added here)

License

MIT License