Run and interact with MATLAB using AI applications through the Model Context Protocol, enabling AI agents to execute MATLAB code, manage sessions, and assess code quality.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
Run and interact with MATLAB using AI applications through the Model Context Protocol, enabling AI agents to execute MATLAB code, manage sessions, and assess code quality.
The MATLAB MCP Server is the official MathWorks server that enables AI applications to start MATLAB, execute MATLAB code, and assess code quality through the Model Context Protocol. It supports multiple session modes, custom working directories, and both desktop and headless MATLAB operation.
Installation
Prerequisites
MATLAB R2021a or later installed and added to system PATH
The server supports MATLAB releases from the past five years
For Claude Code
# Basic installation
claude mcp add --transport stdio matlab -- /path/to/matlab-mcp-server
# With custom working folder
claude mcp add --transport stdio matlab -- /path/to/matlab-mcp-server --initial-working-folder=/home/user/project
# With nodesktop mode
claude mcp add --transport stdio matlab -- /path/to/matlab-mcp-server --matlab-display-mode=nodesktop
For Claude Desktop
Install the Filesystem extension in Claude Desktop (Settings > Extensions > Browse extensions)
Windows:
Download from releases page: matlab-mcp-server-windows-x64.exe
Build from source:
go install github.com/matlab/matlab-mcp-server/cmd/matlab-mcp-server@latest
Configuration Arguments
Command-Line Flags
# Specify MATLAB installation
--matlab-root=/usr/local/MATLAB/R2026a
# Initialize MATLAB immediately on startup
--initialize-matlab-on-startup=true# Set working directory
--initial-working-folder=/home/user/myproject
# Run without MATLAB desktop
--matlab-display-mode=nodesktop
# Session modes
--matlab-session-mode=new # Always start new MATLAB
--matlab-session-mode=auto # Connect to existing or start new (default)
--matlab-session-mode=existing # Only connect to existing MATLAB
Environment Variables
# Equivalent to --matlab-rootexport MW_MCP_SERVER_MATLAB_ROOT=/usr/local/MATLAB/R2026a
# Equivalent to --initial-working-folderexport MW_MCP_SERVER_INITIAL_WORKING_FOLDER=/home/user/project
# Equivalent to --matlab-display-modeexport MW_MCP_SERVER_MATLAB_DISPLAY_MODE=nodesktop
# Equivalent to --matlab-session-modeexport MW_MCP_SERVER_MATLAB_SESSION_MODE=existing
Using Existing MATLAB Sessions
For MATLAB R2023a and later:
First-time setup:
./matlab-mcp-server --setup-matlab
This installs the MATLAB MCP Server Toolbox.
In MATLAB command window:
shareMATLABSession()
Add this to your MATLAB startup.m for automatic sharing:
% In startup.m
shareMATLABSession()
Configure MCP server:
claude mcp add --transport stdio matlab -- /path/to/matlab-mcp-server --matlab-session-mode=existing
MCP Tools Available
The server exposes these tools to AI applications:
% Save data to file
data = rand(100, 3);
save('mydata.mat', 'data');
% Load and process
load('mydata.mat');
mean_values = mean(data);
writematrix(mean_values, 'results.csv');
Plotting and Visualization
% Create and save a plot
x = linspace(0, 2*pi, 100);
y = sin(x);
figure;
plot(x, y);
title('Sine Wave');
xlabel('x');
ylabel('sin(x)');
saveas(gcf, 'sine_plot.png');
Matrix Operations
% Linear algebra operations
A = magic(5);
b = sum(A, 2);
x = A \ b; % Solve Ax = b
% Check solution
residual = norm(A*x - b);
fprintf('Residual: %.2e\n', residual);
Signal Processing
% Generate and filter signal
Fs = 1000; % Sampling frequency
t = 0:1/Fs:1-1/Fs;
signal = sin(2*pi*50*t) + 0.5*randn(size(t));
% Apply low-pass filter
[b, a] = butter(6, 100/(Fs/2));
filtered = filter(b, a, signal);
% Compute FFT
Y = fft(filtered);
P2 = abs(Y/length(filtered));
P1 = P2(1:length(filtered)/2+1);
% Define reusable function
function [mean_val, std_val] = analyzeData(data)
mean_val = mean(data, 'omitnan');
std_val = std(data, 'omitnan');
% Visualize
figure;
histogram(data, 30);
title(sprintf('Mean: %.2f, Std: %.2f', mean_val, std_val));
end
% Use the function
results = rand(1000, 1) * 100;
[m, s] = analyzeData(results);
Simulink Integration
% Load and simulate Simulink model
load_system('mymodel');
simOut = sim('mymodel', 'StopTime', '10');
% Extract and plot results
time = simOut.tout;
output = simOut.yout;
plot(time, output);
Troubleshooting
MATLAB Not Found
Problem: Server cannot locate MATLAB installation.
Solution:
# Explicitly specify MATLAB root
--matlab-root=/Applications/MATLAB_R2026a.app # macOS
--matlab-root=/usr/local/MATLAB/R2026a # Linux
--matlab-root=C:\\Program Files\\MATLAB\\R2026a # Windows# Or set environment variableexport MW_MCP_SERVER_MATLAB_ROOT=/usr/local/MATLAB/R2026a
Connection to Existing Session Fails
Problem: Cannot connect with --matlab-session-mode=existing
Solution:
Ensure MATLAB MCP Server Toolbox is installed:
./matlab-mcp-server --setup-matlab
In MATLAB, run:
shareMATLABSession()
Verify connection status:
status = shareMATLABSession('status')
Path Issues
Problem: MATLAB cannot find scripts or data files.
Solution:
% Check current directory
pwd
% Change directory
cd('/path/to/project')
% Add to path
addpath('/path/to/scripts');
addpath(genpath('/path/to/project')); % Include subdirectories
Graphics/Desktop Issues
Problem: Commands requiring GUI fail in nodesktop mode.
Solution: Graphics commands still work in nodesktop mode, but if issues persist:
# Switch to desktop mode
--matlab-display-mode=desktop
Memory Issues
Problem: Out of memory errors with large datasets.
Solution:
% Clear workspace
clear all
% Close figures
close all
% Use memory-efficient operations
% Instead of loading entire file:
data = load('largefile.mat');
% Use memory mapping:
m = memmapfile('largefile.dat', 'Format', 'double');
Code Execution Timeout
Problem: Long-running code appears to hang.
Solution:
% Add progress indicators
for i = 1:1000
% Process
if mod(i, 100) == 0
fprintf('Progress: %d/1000\n', i);
end
end
% Use parallel processing for large tasks
parfor i = 1:1000
% Parallel computation
end
Version Compatibility
Problem: Functions not available in older MATLAB versions.
Solution:
% Check MATLAB version
ver('MATLAB')
% Conditional code based on version
if verLessThan('matlab', '9.10') % R2021a
warning('Some features require R2021a or later');
end
Advanced Configuration
Custom Tools Extension
Create custom MCP tools by providing a JSON extension file:
# Development with latest MATLAB
claude mcp add --transport stdio matlab-dev -- /path/to/matlab-mcp-server --matlab-root=/usr/local/MATLAB/R2026a
# Production with stable MATLAB
claude mcp add --transport stdio matlab-prod -- /path/to/matlab-mcp-server --matlab-root=/usr/local/MATLAB/R2024b
Project-Specific Configuration
For VS Code, use workspace-specific .vscode/mcp.json:
