The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the MATLAB MCP Server listing page.
English • Español • 日本語 • 한국어 • 简体中文
Run MATLAB® using AI applications with the official MATLAB MCP Server from MathWorks®. The MATLAB MCP Server allows your AI applications to:
To assist your agent in using MATLAB and Simulink, you can use skills from MATLAB Agentic Toolkit (GitHub) and Simulink Agentic Toolkit (GitHub), which can also install this MCP server for you.
Install MATLAB (MathWorks) R2021a or later and add it to the system PATH.
The server supports any AI application that uses Model Context Protocol. To set up the MATLAB MCP Server for Claude Desktop, skip to the instructions for Claude Desktop. To set up the server for other applications, follow these instructions:
For Windows or Linux, Download the Latest Release. (Alternatively, you can build from source: install Go and build the binary using go install github.com/matlab/matlab-mcp-server/cmd/matlab-mcp-server@latest).
For macOS, first download the latest release by running the following command in your terminal:
Then grant executable permissions to the downloaded binary so you can run the MATLAB MCP Server:
Add the MATLAB MCP Server to your AI application. You can find instructions for adding MCP servers in the documentation of your AI application. For example instructions on using Claude Code®, Claude Desktop®, and GitHub Copilot in Visual Studio® Code, see below. Note that you can customize the server by specifying optional Arguments.
In your terminal, run the following, remembering to insert the full path to the server binary you acquired in the setup:
You can customize the server by specifying optional Arguments. Note the -- separator between Claude Code's options and the server arguments:
For details on adding MCP servers in Claude Code, see Add a local stdio server (Claude Code). To remove the server later, run:
You install the MATLAB MCP Server in Claude Desktop using the MATLAB MCP Server bundle.
Install the Filesystem extension in Claude Desktop to allow Claude to read and write files on your system. In Claude Desktop, click Settings > Extensions > Browse extensions. Search for the Filesystem extension developed by Anthropic and click Install. Specify the folders you want to allow the MCP server to access, then toggle the Disabled button to Enable the Filesystem extension.
Download the MATLAB MCP Server bundle matlab-mcp-server.mcpb from the Latest Release page.
To install the MATLAB MCP Server bundle as a desktop extension, double click the downloaded matlab-mcp-server.mcpb file and click Install in Claude Desktop. (Alternatively, navigate in Claude to File menu > Settings > Extensions > Advanced Settings > Install Extension and select the matlab-mcp-server.mcpb file. Click Install).
To customize the behaviour of the MATLAB MCP Server, navigate to Settings > Extensions > Configure, where you can modify the server's Arguments.
In your VS Code workspace, create a file named .vscode/mcp.json. Insert the following JSON, remembering to specify the full path to the server binary you acquired in the setup, as well as any Arguments. Then save the file. (Note that on Windows, your paths require extra slashes as escape characters).
For more information about using MCP servers in VS Code, see Add and Manage MCP servers in VS Code (VS Code).
In your terminal, run the following, remembering to insert the full path to the server binary you acquired in the setup:
You can customize the server by specifying optional Arguments. For example:
[!NOTE] Windows users: Codex does not forward the
WINDIRenvironment variable, which you need to run MATLAB and Simulink (see #32).To add the variable, open your Codex config file at
C:\Users\<username>\.codex\config.toml, and addenv_vars = ["WINDIR"]to the[mcp_servers.matlab]section:
Customize the behavior of the server by specifying arguments in one of these ways:
.json file).MW_MCP_SERVER_, convert to uppercase, and replace hyphens (-) with underscores (_). For example, the argument --matlab-root becomes the environment variable MW_MCP_SERVER_MATLAB_ROOT. CLI flags take precedence over environment variables, if you use both.| Argument | Description | Example |
|---|---|---|
| help | Displays help information for all arguments. | --help |
| version | Displays the version of the MATLAB MCP Server. | --version |
| matlab-root | Full path specifying which MATLAB to start. Do not include /bin in the path. By default, the server tries to find the first MATLAB on the system PATH. | Windows: --matlab-root=C:\\Program Files\\MATLAB\\R2026a Linux/macOS: --matlab-root=/home/usr/MATLAB/R2026aAs an environment variable: MW_MCP_SERVER_MATLAB_ROOT=/home/usr/MATLAB/R2026a |
| initialize-matlab-on-startup | To initialize MATLAB as soon as you start the server, set this argument to true. By default, MATLAB only starts when the first tool is called. | --initialize-matlab-on-startup=true |
| initial-working-folder | Specify the folder where MATLAB starts. If you do not specify a value, MATLAB starts at the path of your AI application's first Root (MCP). If you have not defined a root, MATLAB starts in these locations:
| Windows: --initial-working-folder=C:\\Users\\username\\MyProject Linux/macOS: --initial-working-folder=/Users/username/MyProject |
| matlab-display-mode | Specify whether to show the MATLAB desktop. Use desktop mode (default) to show the MATLAB desktop. Use nodesktop mode to use MATLAB only from your AI application, without the MATLAB desktop. Note that in nodesktop mode, commands requiring a graphical interface (such as edit, open, open_system, uifigure, and appdesigner) will still open MATLAB windows on your desktop. | --matlab-display-mode=nodesktop |
| matlab-session-mode | Specify whether the MCP server starts a new MATLAB or connects to an existing MATLAB session (supported for MATLAB R2023a onwards). The default is auto mode.new mode: The MCP server starts a new MATLAB session. auto mode (default): The server tries to connect to an existing MATLAB session, which you must have configured for existing mode using the instructions below. If the server is unable to find an existing MATLAB session, it starts a new one. existing mode: The server tries to connect to an existing MATLAB session. Do not use this mode in conjunction with the arguments matlab-root, initial-working-folder, or matlab-display-mode. You must have configured your MATLAB session beforehand to use this mode, with these steps:
| --matlab-session-mode=existing |
| extension-file | To use custom MCP tools, provide a path to a JSON file that defines your tools. You can also use multiple extension files. For details on using custom tools, see Use Custom Tools with the MATLAB MCP Server. | Windows: --extension-file=C:\\Users\\name\\my-tools.json Linux/macOS: --extension-file=/path/to/my-tools.json Using multiple extension files: Windows: --extension-file=C:\\path\\to\\tools-1.json --extension-file=C:\\path\\to\\tools-2.jsonLinux/macOS: --extension-file=/path/to/tools1.json --extension-file=/path/to/tools2.json Using environment variables: Windows: MW_MCP_SERVER_EXTENSION_FILE=C:\Users\name\tools1.json;C:\Users\name\tools2.json Linux/macOS: MW_MCP_SERVER_EXTENSION_FILE=/path/to/tools1.json:/path/to/tools2.json |
| log-folder | Specify the folder where the MCP server stores log files. If not specified, the server uses the default temporary folder of your operating system. | Windows: --log-folder=C:\\Users\\name\\AppData\\Local\\Temp Linux/macOS: --log-folder=/tmp/my-logs |
| log-level | The log levels of the MCP server. Valid values, in order of decreasing verbosity, are debug, info, warn, and error. The default is info. | --log-level=debug |
| disable-telemetry | To disable anonymized data collection, set this argument to true. For details, see Data Collection. | --disable-telemetry=true |
detect_matlab_toolboxes
check_matlab_code
script_path (string): Absolute path to the MATLAB script file to analyze. Must be a valid .m file. The file is not modified during analysis. Example: C:\Users\username\matlab\myFunction.m or /home/user/scripts/analysis.m.evaluate_matlab_code
code (string): MATLAB code to evaluate.project_path (string): Absolute path to your project directory. MATLAB sets this directory as the current working folder. Example: C:\Users\username\matlab-project or /home/user/research.run_matlab_file
.m file.script_path (string): Absolute path to the MATLAB script file to execute. Must be a valid .m file. Example: C:\Users\username\projects\analysis.m or /home/user/matlab/simulation.m.run_matlab_test_file
script_path (string): Absolute path to the MATLAB test script file. Must be a valid .m file containing MATLAB unit tests. Example: C:\Users\username\tests\testMyFunction.m or /home/user/matlab/tests/test_analysis.m.The MCP server provides Resources (MCP) to help your AI application write MATLAB code. To see instructions for using this resource, refer to the documentation of your AI application that explains how to use resources.
matlab_coding_guidelines
guidelines://codingtext/markdownplain_text_live_code_guidelines
.m file format, suitable for version control and AI-assisted development. Note that to run plain text live scripts you need MATLAB R2025a or newer. For details, see Live Code File Format (MathWorks).guidelines://plain-text-live-codetext/markdownThe MATLAB MCP Server may collect fully anonymized information about your usage of the server and send it to MathWorks. This data collection helps MathWorks improve products and is on by default. To opt out of data collection, set the argument --disable-telemetry to true.
When using the MATLAB MCP Server, you should thoroughly review and validate all tool calls before you run them. Always keep a human in the loop for important actions and only proceed once you are confident the call will do exactly what you expect. For more information, see User Interaction Model (MCP) and Security Considerations (MCP).
The license is available in the LICENSE.md file in this GitHub repository.
MCP servers are only permitted to be used with MATLAB in accordance with the MathWorks Software License Agreement, and must not be shared by multiple users. Contact MathWorks if you need to support shared or centralized server use.
MathWorks encourages you to use this repository and provide feedback. To request technical support or submit an enhancement request, create a GitHub issue or contact MathWorks Technical Support.
Copyright 2025-2026 The MathWorks, Inc.