MATLAB MCP Integration

Created By

jigarbhoye048 months ago

MCP server to run MATLAB code from LLM via the Matlab Engine API.

Content

MATLAB MCP Integration

This is an implementation of a Model Context Protocol (MCP) server for MATLAB. It allows MCP clients (like LLM agents or Claude Desktop) to interact with a shared MATLAB session using the MATLAB Engine API for Python.

Features

Execute MATLAB Code: Run arbitrary MATLAB code snippets via the runMatlabCode tool.
Retrieve Variables: Get the value of variables from the MATLAB workspace using the getVariable tool.
Structured Communication: Tools return results and errors as structured JSON for easier programmatic use by clients.
Non-Blocking Execution: MATLAB engine calls are run asynchronously using asyncio.to_thread to prevent blocking the server.
Standard Logging: Uses Python's standard logging module, outputting to stderr for visibility in client logs.
Shared Session: Connects to an existing shared MATLAB session.

TODO:

Add a setVariable tool to write data to the MATLAB workspace.
Add a runScript tool to execute .m files directly.
Add tools for workspace management (e.g., clearWorkspace, getWorkspaceVariables).
Expand matlab_to_python helper to handle more complex data types (structs, cell arrays, objects).
Add support for interacting with Simulink models.

Requirements

Python 3.12 or higher
MATLAB (R2023a or higher recommended - check MATLAB Engine API for Python compatibility) with the MATLAB Engine API for Python installed.
numpy Python package.

Installation

Clone this repository:

git clone https://github.com/jigarbhoye04/MatlabMCP.git
cd MatlabMCP

Set up a Python virtual environment (recommended):

# Install uv if you haven't already: https://github.com/astral-sh/uv
uv init
uv venv
source .venv/bin/activate  # On Windows use: .venv\Scripts\activate

Install dependencies:
```
uv pip sync
```
Ensure MATLAB is installed and the MATLAB Engine API for Python is configured for your Python environment. See MATLAB Documentation.
Start MATLAB and share its engine: Run the following command in the MATLAB Command Window:
```
matlab.engine.shareEngine
```
You can verify it's shared by running matlab.engine.isEngineShared in MATLAB (it should return true or 1). The MCP server needs this shared engine to connect.

Configuration (for Claude Desktop)

To use this server with Claude Desktop:

Go to Claude Desktop -> Settings -> Developer -> Edit Config.

This will open claude_desktop_config.json. Add or modify the mcpServers section to include the MatlabMCP configuration:

{
  "mcpServers": {
    "MatlabMCP": {
      "command": "C:\\Users\\username\\.local\\bin\\uv.exe", // Path to your uv executable
      "args": [
        "--directory",
        "C:\\Users\\username\\Desktop\\MatlabMCP\\", // ABSOLUTE path to the cloned repository directory
        "run",
        "main.py"
      ]
      // Optional: Add environment variables if needed
      // "env": {
      //   "MY_VAR": "value"
      // }
    }
    // Add other MCP servers here if you have them
  }
}

IMPORTANT: Replace C:\\Users\\username\\... paths with the correct absolute paths for your system.
Save the file and restart Claude Desktop.
Logging: Server logs (from Python's logging module) will appear in Claude Desktop's MCP log files (accessible via tail -f ~/Library/Logs/Claude/mcp-server-MatlabMCP.log on macOS or checking %APPDATA%\Claude\logs\ on Windows).

Development

Project Structure:

MatlabMCP/
├── .venv/                     # Virtual environment created by uv
├── Docs/
│   └── Images/
│   └── Updates.md             # Documentation for updates and changes
├── main.py                    # The MCP server script
├── pyproject.toml             # Project metadata and dependencies
├── README.md                  # This file
└── uv.lock                    # Lock file for dependencies