Examples

This page contains practical, real-world usage examples for the MATLAB MCP Server across JavaScript, MATLAB, and Python, progressing from simple to advanced scenarios.

Basic MATLAB Execution

Example 1: Simple Calculation

What it demonstrates: Basic code execution and output capture.

Prerequisites:

MATLAB R2022b+ installed
Server running: matlab-mcp --inspect (inspect mode skips pool startup)

% Calculate eigenvalues of a 3×3 magic square
A = magic(3);
eigenvalues = eig(A);
disp(eigenvalues)

Expected output:

   15.0000
    4.1231
   -4.1231

Example 2: Matrix Operations

What it demonstrates: Larger computations and result formatting.

% Create two 100×100 random matrices, multiply, show trace
A = rand(100);
B = rand(100);
C = A * B;
fprintf('Trace of A*B: %.6f\n', trace(C));
fprintf('Matrix dimensions: %d×%d\n', size(C, 1), size(C, 2));

Expected output:

Trace of A*B: 4982.731405
Matrix dimensions: 100×100

Example 3: Solve Linear System

What it demonstrates: Structured computation and array output.

% Solve: 3x + 2y - z = 1
%        2x - 2y + 4z = -2
%        -x + 0.5y - z = 0
A = [3 2 -1; 2 -2 4; -1 0.5 -1];
b = [1; -2; 0];
x = A \ b;
fprintf('Solution: x=%.4f, y=%.4f, z=%.4f\n', x(1), x(2), x(3));

Plotting and Visualization

All figures are automatically converted to interactive Plotly JSON and static PNG thumbnails.

Example 4: Simple Line Plot

What it demonstrates: Figure creation and automatic Plotly conversion.

% Plot sin(x) from 0 to 2π
x = linspace(0, 2*pi, 200);
y = sin(x);
plot(x, y, 'LineWidth', 2, 'DisplayName', 'sin(x)');
xlabel('x');
ylabel('y');
title('Sine Wave');
grid on;
legend;

Agent sees:

Interactive Plotly chart (zoom, pan, hover)
Static PNG thumbnail
Axis labels and title preserved

graph LR
    A["MATLAB figure<br/>plot(x,y)"] -->|mcp_extract_props.m| B["Figure JSON<br/>properties"]
    B -->|Plotly mapper| C["Interactive<br/>Plotly JSON"]
    C -->|Browser| D["Agent sees<br/>interactive chart"]

Example 5: 3D Surface Plot

What it demonstrates: Complex 3D visualization.

% Peaks function as 3D surface
[X, Y] = meshgrid(-3:0.1:3);
Z = peaks(X, Y);
surf(X, Y, Z);
colorbar;
shading interp;
title('Peaks Function');
xlabel('X');
ylabel('Y');
zlabel('Z');

Example 6: Subplot Layout

What it demonstrates: Multi-panel figures with subplot domain computation.

% Four frequency sine waves in 2×2 grid
t = linspace(0, 1, 1000);
freqs = [5 10 20 50];

for i = 1:4
    subplot(2, 2, i);
    plot(t, sin(2*pi*freqs(i)*t), 'LineWidth', 1.5);
    title(sprintf('%d Hz', freqs(i)));
    xlabel('Time (s)');
    ylabel('Amplitude');
    grid on;
end

Agent receives:

Four traces in proper subplot domains
Each subplot as a separate Plotly subplot with correct axis positioning

Signal Processing

Example 7: FFT Analysis

What it demonstrates: Signal processing with Plotly dual-subplot visualization.

% Generate 440 Hz signal with noise, compute FFT
fs = 8000;              % Sample rate
t = 0:1/fs:0.5;        % 0.5 second duration
signal = sin(2*pi*440*t) + 0.3*randn(size(t));

% FFT analysis
N = length(signal);
Y = fft(signal);
f = (0:N-1) * fs / N;

% Plot results
figure;
subplot(2, 1, 1);
plot(t*1000, signal);
title('Time Domain: 440 Hz + Noise');
xlabel('Time (ms)');
ylabel('Amplitude');
grid on;

subplot(2, 1, 2);
plot(f(1:N/2), abs(Y(1:N/2))/N);
title('Frequency Domain (FFT)');
xlabel('Frequency (Hz)');
ylabel('Magnitude');
xlim([0 2000]);
grid on;

Example 8: Filter Design and Response

What it demonstrates: Signal Processing Toolbox integration.

% Design low-pass Butterworth filter
fc = 1000;          % Cutoff frequency (Hz)
fs = 8000;          % Sample rate (Hz)
[b, a] = butter(4, fc/(fs/2));  % 4th order

% Plot frequency response
freqz(b, a, 1024, fs);
title('Low-Pass Butterworth Filter (1 kHz cutoff)');
grid on;

Asynchronous Jobs (Long-Running)

Code exceeding sync_timeout (default 30s) automatically promotes to async. Report progress with mcp_progress().

Example 9: Monte Carlo Simulation

What it demonstrates: Automatic async promotion and progress tracking.

% Estimate π via Monte Carlo (1 million trials)
n_trials = 1e6;
inside_circle = 0;

for i = 1:n_trials
    x = rand();
    y = rand();
    if x^2 + y^2 <= 1
        inside_circle = inside_circle + 1;
    end
    
    % Report progress every 100k trials
    if mod(i, 1e5) == 0
        pct = (i / n_trials) * 100;
        msg = sprintf('Trial %d / %d (%.1f%%)', i, n_trials, pct);
        mcp_progress(__mcp_job_id__, pct, msg);
    end
end

pi_estimate = 4 * inside_circle / n_trials;
fprintf('\nEstimated π = %.6f (error: %.6f)\n', ...
    pi_estimate, abs(pi_estimate - pi));

Agent experience:

Receives job ID immediately
Polls progress: "Trial 100000 / 1000000 (10%)" → "Trial 500000 / 1000000 (50%)" → etc.
Gets final result: "Estimated π = 3.141592"

Example 10: Large SVD Computation

What it demonstrates: Async promotion for matrix operations.

% Singular value decomposition of large matrix
% This ~45 second computation triggers automatic async
A = randn(5000, 5000);

fprintf('Computing SVD of 5000×5000 matrix...\n');
[U, S, V] = svd(A, 'econ');

fprintf('Singular values (top 10):\n');
disp(diag(S)(1:10));
fprintf('Computation complete.\n');

File Operations

Example 11: Save and Read Data

What it demonstrates: Persisting results and agent file retrieval.

% Generate and save data
x = linspace(0, 10, 100);
y = sin(x) .* exp(-x/5);

% Save as MAT file (binary)
save('signal_data.mat', 'x', 'y');

% Also save as CSV (human-readable)
T = table(x', y', 'VariableNames', {'x', 'y'});
writetable(T, 'signal_data.csv');

fprintf('Saved signal_data.mat and signal_data.csv\n');

Agent can then:

read_data(filename="signal_data.mat", format="summary")

Returns:

Name      Size       Bytes  Class
x         1x100       800   double array
y         1x100       800   double array

Or:

read_data(filename="signal_data.csv", format="summary")

Returns CSV text inline.

Example 12: Generate and Read Plot

What it demonstrates: Saving figures and agent image retrieval.

% Create and save a plot
figure('Visible', 'off');
x = linspace(-pi, pi, 200);
plot(x, sin(x), 'b-', 'LineWidth', 2);
plot(x, cos(x), 'r--', 'LineWidth', 2);
legend('sin(x)', 'cos(x)');
xlabel('x');
ylabel('y');
title('Trigonometric Functions');
grid on;

% Save as PNG
saveas(gcf, 'trig_functions.png');
close;
fprintf('Saved trig_functions.png\n');

Agent retrieves:

read_image(filename="trig_functions.png")

Image displays inline in Claude Desktop, Cursor, etc.

Code Quality & Validation

Example 13: Check Code Quality

What it demonstrates: Static MATLAB linting via MCP tool.

% Code with potential issues (unused variables, implicit expansion)
function result = analyze_data(data)
    mean_val = mean(data);
    unused_var = sum(data);  % Unused
    
    % Implicit expansion (potential issue)
    result = data * [1 2 3];  % Broadcasting
end

Agent calls:

check_code(code=<<< function result = analyze_data(data) ... end >>>)

Returns warnings:

Line 4: Implicit expansion occurs for operator *.
Line 3: Variable "unused_var" is not used.

Example 14: Custom Tool Example

What it demonstrates: Using agent-defined custom MATLAB tools.

Assuming custom_tools.yaml contains:

tools:
  - name: signal_analysis
    function: signal_analysis
    parameters:
      - name: signal
        type: array
        description: Input signal vector
      - name: fs
        type: number
        description: Sample rate in Hz

And corresponding MATLAB function:

function [freq_peaks, power] = signal_analysis(signal, fs)
    Y = fft(signal);
    power = abs(Y).^2 / length(signal);
    [~, peak_idx] = max(power);
    freq_peaks = (peak_idx - 1) * fs / length(signal);
end

Agent uses it like any built-in tool:

signal_analysis(signal=[values...], fs=8000)

Monitoring & Diagnostics

Example 15: Check Server Health

What it demonstrates: Real-time metrics access.

// JavaScript / Node.js example
async function getServerHealth() {
  const response = await fetch('http://127.0.0.1:8765/health', {
    headers: {
      'Authorization': 'Bearer YOUR_TOKEN_HERE'
    }
  });
  
  const health = await response.json();
  console.log('Server Status:', health.status);
  console.log('Uptime:', health.uptime_seconds);
  console.log('Engines:', health.engine_count);
  console.log('Active Jobs:', health.active_jobs);
  
  return health;
}

getServerHealth();

Example 16: Monitor Pool Utilization

What it demonstrates: Pool metrics via MCP tool.

Agent calls:

get_pool_status()

Returns:

{
  "total_engines": 8,
  "busy_engines": 6,
  "available_engines": 2,
  "max_engines": 10,
  "utilization_percent": 75.0
}

Configuration & Deployment

Configuration 1: Single-User Local Setup

# config.yaml for local development
server:
  transport: "stdio"
  log_level: "debug"

pool:
  min_engines: 1
  max_engines: 2

execution:
  sync_timeout: 30

security:
  blocked_functions_enabled: true

Run:

matlab-mcp --config config.yaml

Configuration 2: Multi-User Server with Auth

# config.yaml for production multi-user
server:
  transport: "streamablehttp"
  host: "127.0.0.1"
  port: 8765
  log_level: "info"

pool:
  min_engines: 4
  max_engines: 16
  warmup_percent: 50

execution:
  sync_timeout: 60
  max_concurrent_jobs: 100

sessions:
  max_sessions: 50
  session_timeout: 3600

security:
  blocked_functions_enabled: true
  require_proxy_auth: false

monitoring:
  enabled: true
  metrics_db: "./monitoring/metrics.db"
  dashboard_port: 8765

Run with auth:

export MATLAB_MCP_AUTH_TOKEN=your-secure-token-here
matlab-mcp --config config.yaml

Architecture Diagram

graph TB
    Agent["AI Agent<br/>(Claude, Cursor, etc.)"]
    Transport["Transport Layer<br/>(stdio, HTTP, SSE)"]
    Auth["Auth Middleware<br/>(Bearer Token)"]
    MCP["MCP Tool Handler<br/>(execute_code, etc.)"]
    Sec["Security Validator<br/>(blocked functions)"]
    Job["Job Executor<br/>(async promotion)"]
    Pool["Engine Pool<br/>(elastic scaling)"]
    MATLAB["MATLAB Engine<br/>(R2022b+)"]
    Fmt["Result Formatter<br/>(Plotly, PNG)"]
    
    Agent -->|Request| Transport
    Transport -->|MCP JSON-RPC| Auth
    Auth -->|Validated| MCP
    MCP -->|Check code| Sec
    Sec -->|Valid| Job
    Job -->|Acquire engine| Pool
    Pool -->|Execute| MATLAB
    MATLAB -->|Output| Job
    Job -->|Format| Fmt
    Fmt -->|Response| Agent

Tips & Best Practices

Use progress reporting for jobs over 10 seconds — agents appreciate visibility:
```
mcp_progress(__mcp_job_id__, percent, message);
```

Vectorize when possible — MATLAB is fastest with array operations, slowest with loops:

% ✓ Fast
result = sin(x) .* cos(y);

% ✗ Slow
for i = 1:length(x)
  result(i) = sin(x(i)) * cos(y(i));
end

Close figures explicitly to avoid memory leaks:

figure; plot(...); saveas(gcf, 'output.png'); close;

Validate inputs in custom tools — agents may send unexpected types:

function result = my_tool(data)
  if ~isvector(data), error('Input must be a vector'); end
  result = sum(data);
end

Use inspect mode for testing without MATLAB:

matlab-mcp --inspect  # Pool disabled, tools return mocks

Examples

Examples

Basic MATLAB Execution

Example 1: Simple Calculation

Example 2: Matrix Operations

Example 3: Solve Linear System

Plotting and Visualization

Example 4: Simple Line Plot

Example 5: 3D Surface Plot

Example 6: Subplot Layout

Signal Processing

Example 7: FFT Analysis

Example 8: Filter Design and Response

Asynchronous Jobs (Long-Running)

Example 9: Monte Carlo Simulation

Example 10: Large SVD Computation

File Operations

Example 11: Save and Read Data

Example 12: Generate and Read Plot

Code Quality & Validation

Example 13: Check Code Quality

Example 14: Custom Tool Example

Monitoring & Diagnostics

Example 15: Check Server Health

Example 16: Monitor Pool Utilization

Configuration & Deployment

Configuration 1: Single-User Local Setup

Configuration 2: Multi-User Server with Auth

Architecture Diagram

Tips & Best Practices

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Clone this wiki locally