Crawl time (ISO local): 2026-05-18 05:17:43

Foreword

This article is compiled from a real Windows environment configuration process, for readers who have installed * * MATLAB/Simulink * * and * * Python * *, and want to let AI directly operate the Simulink model through * * Model Context Protocol (MCP) * * in * * Cursor or other Agents * *.
Upstream open source project: * * [sohumsuthar/simulink-mcp] (https://github.com/sohumsuthar/simulink-mcp) * * (license: * * PolyForm Noncommercial 1.0.0 * *, commercial use please evaluate yourself).

# # I. What you will get

After the configuration is complete, the AI in the Cursor can call about * * 14 structured tools * * through the MCP, for example:

• Load/Close/Save .slx Model
• List modules, read and write parameters, add and remove modules, connectivity
• Read and setup solver and model configurations
• Run the simulation and return the image or data

MATLAB uses * * lazy start * *: MCP shakes hands quickly, and the MATLAB engine is started only when * * the tool is actually called for the first time * *. Cold start often takes * * about 15 to 20 seconds * *, which is normal.

After configuring this mcp, you don’t need to drag and set it frequently in Simulink. You just need to use natural language and AI to say what kind of model you want to build and what kind of loop will come out. AI will help you draw it directly with the fastest speed.

# # II. environment and version: the easiest chain TO step ON

The whole link has to satisfy three things at the same time:

1. • • MATLAB Engine for Python * * supports your choice of * * Python Minor * * (as determined by the * * MATLAB Distribution * *).
2. • • Official Python package mcp (mcp [cli] > = 1.2.0) * * Requirements * * Python ≥ 3.10 * *.
3. • • simulink-mcp * * relies on the above mcp, so the interpreter for * * actually running MCP must be 3.10 + * * and the same as the one installed with Engine * * *.

# # # Typical Conflict (Real Case)

• Engine * * setup.py included with older versions of MATLAB (e.g. * * R2022a * *) only declares support for Python 2.7/3 .7/3.8/3 .9 * *, execute pip install on * * Python 3.10 * *.`will be reported directly to version not supported.
• If you insist on using * * Python 3.9 * * with the Engine, * * mcp > = 1.2.0 cannot be installed * * (Python ≥ 3.10 is required for this package on PyPI).

• • Conclusion: * * To use the current mainline * * simulink-mcp * *, you generally need * * newer MATLAB * * (the official README is mostly tested for environments such as R2024a/R2024b) + * * Python 3.10 or 3.11 * * (specific to the MathWorks documentation for “MATLAB Engine for Python”, and avoid using newer versions that are not yet supported by the Engine, such as Python 3.13 in some environments).

# # III. Confirming the MATLAB root directory

When MATLAB is ready to be opened, execute on the * * command line * *:

matlabroot

Note the output path (e.g. F:\ MATLAB2024). The Engine source directory is:

<matlabroot>\extern\engines\python

Expected * * setup.py * * under this directory.Open the path in the explorer so that you can cd it later in PowerShell.! [Insert image description here] (./1.png)

# # IV. Installing the MATLAB Engine for Python

# # # 4.1 Unique Python interpreter selected

After * * Cursor MCP’s command * *, pip install, import matlab.engine self-test, * * must * * point to * * the same * * python.exe (it is recommended to always use * * absolute path * *, avoid mixing multiple python in path).

# # # 4.2 Recommended: Execute setup.py install in the Engine directory

In * * PowerShell * * (Note: when calling an executable with a path, precede it with * * & * *):

cd "<matlabroot>\extern\engines\python"
& "C:\Path\To\Python310\python.exe" setup.py install

• • Why not just pip install.? * *
    pip sometimes copies the source code to a temporary directory to rebuild; in some environments, false positives such as * * Installation corruption * * are triggered.* * setup.py install' * * directly under * * extern\ engines\ python` source directory * *, tends to be more stable.

# # # 4.3 Verifying the Engine

& "C:\Path\To\Python310\python.exe" -c "import matlab.engine; print('OK')"

OK should be output.

# # V. Installing simulink-mcp

# # # 5.1 PyPI Package Name Description

pip install simulink-mcp may appear in documents or old notes, but * * there may not be a package with the same name on PyPI * * (or it may not match your desired repository).* * Reliable way * * is to install locally after cloning from * * GitHub * *:
You can also download the entire archive directly on github:https://github.com/sohumsuthar/simulink-mcp

! [insert image description here] (./2.png)

cd C:\Path\To\simulink-mcp

# # # 5.2 Syntax for PowerShell calling pip

& "C:\Path\To\Python310\python.exe" -m pip install .

# # # 5.3 Dependency build with mcp: step-by-step installation recommended

Upgrade the build chain first and install * * mcp [cli] * * (need to be able to access PyPI; if in the country, you can change the image, but when the image is not synchronized, “Package not found” will appear):

Remove-Item Env:HTTP_PROXY,Env:HTTPS_PROXY,Env:ALL_PROXY,Env:http_proxy,Env:https_proxy,Env:all_proxy -ErrorAction SilentlyContinue

cd "C:\Path\To\simulink-mcp"
& "C:\Path\To\Python310\python.exe" -m pip install --upgrade pip setuptools wheel hatchling "mcp[cli]>=1.2.0" --index-url https://pypi.org/simple --trusted-host pypi.org --trusted-host files.pythonhosted.org

Reinstall this project (* * --no-build-isolation * * reduces the probability of failure of subprocesses to repeatedly pull hatchling):

& "C:\Path\To\Python310\python.exe" -m pip install . --no-build-isolation --index-url https://pypi.org/simple --trusted-host pypi.org --trusted-host files.pythonhosted.org

# # # 5.4 Network and proxy issues (real pitfalls)

In case of * * ProxyError (' Cannot connect to proxy. ', FileNotFoundError (...) * *:

• May be related to the * * VPN/system proxy/environment variable HTTP_proxy/HTTPS_proxy * *; try again in the newly opened PowerShell after * * turning off VPN * *.
• Clears the current session proxy variable (see Remove-Item Env:... in the previous section).
• Check * * netsh winhttp show proxy * * and * * netsh winhttp reset proxy * * if necessary.
• Check if * * pip config list * * is configured with an invalid proxy.

If * * SSLError/Tsinghua mirror does not pull hatchling * *: you can temporarily switch to * * official PyPI * * and cooperate with * * ‘—trusted-host` * * (see the order in the previous section).

# # # 5.5 Tips for packing in the same Python as the Family Keg

If the Python also contains * * TensorFlow, Streamlit * *, etc., pip may prompt * * dependent version conflicts * *. For * * simulink-mcp * *, it can still run in most cases; if other projects are abnormal, it is recommended to build a * * separate venv * * for the MCP in the future.

# # VI. Configuring MCP in Cursor

# # # 6.1 Working directory environment variables

Recommended as a folder dedicated to .slx, for example F:/SimulinkModels (forward slash or escape backslash available in JSON).

# # # 6.2 User-level mcp.json example

! [insert image description here] (./3.png)! [Insert image description here] (./4.png)

The path is typically: * * % USERPROFILE %\ .cursor\ mcp.json * *.

“ `text { “mcpServers”: { “simulink”: { “command”: “C:/Users/Your username/AppData/Local/Programs/Python/Python310/python.exe”, “args”: [“-m”, “simulink_mcp”], “env”: { “Simulink_MCP_workdir”: “F:/SimulinkModels” } } } }

! [insert image description here] (./5.png)

MAIN POINTS

- * * `command` * * Always use * * the same `python.exe` absolute path as when installing Engine, `simulink-mcp' * *.
- * * Each MCP server is an independent process * *; changing only `command` of `simulink` * * does not * * automatically change other MCPs (such as services that still write `python` run scripts).
- After the modification, * * fully restart Cursor or reload the window * * and confirm in * * Settings → MCP * * that * * simulink * * is green and the number of tools matches the documentation (e.g. * * 14 tools * *).!
[Insert image description here] (./6.png)

---

# # # VII. Self-inspection checklist

| Checks | Actions |
| --- | --- |
| Minor version of Python | `python -V`, consistent with Engine/`mcp` requirements |
| Engine | `python -c "import matlab.engine; print (' OK ')"` |
| simulink-mcp | `pip show simulink-mcp` |
| MCP interpreter is consistent | `command` in `mcp.json` is the same as` python.exe `above |
| Working Directory | `Simulink_MCP_WORKDIR` is set and the folder exists |
| Licensing | Non-Commercial Use in accordance with PolyForm NC |

---

# # # VIII. Tips for first time use and troubleshooting

- * * First tool call * * before launching MATLAB, please be patient.
- If the MCP reports an error: look at the Cursor * * MCP Log * *; the Task Manager confirms if there is a * * MATLAB * * process.
- Upstream README mentions that invalid Simulink operations may cause the session to crash and the MCP can be * * restarted * *.
- Some versions have initialization requirements for modules with masks such as * * PID * *, and parameters such as saturation and initial value need to be set according to the upstream instructions.

---

# # # IX. Test Examples

* * Prefix: * * Confirm * * Cursor → MCP * * in * * simulink * * is on; `.slx' under `Simulink_MCP_workdir` (or change to your absolute path in the following sentence). In the window of * * Agent mode * * and this dialog * * Simulink MCP tool * * is enabled, copy the following three * * whole paragraphs * * to the input box (can be executed in order ①→ ②→ ③; ② A new model will be created without having to have a file first).

---

# # # # ① See what modules are available and explain their relationship

`` `text
Please use Simulink MCP: load F:/SimulinkModels/1.slx (if it does not exist, load any .slx in my working directory, the path can be changed to my actual file).Then list all the modules of the model under search_depth = 2.Finally, explain in words: what role each module plays in the signal flow, how the negative feedback loop is connected between the reference input/error/control amount/controlled output (as described by the wiring logic you see).

• • Expected Effect: * * Back to Module Path List + a text description of a * * Signal Flow/Feedback Relationship * *.! [Insert image description here] (./7.png)

# # # ② Use the base module to set up a PID controller (PID Controller off-the-shelf module is prohibited)

“ `text Please use Simulink MCP to create and save a new model demo_pid_blocks (under Simulink_MCP_WORKDIR, file name demo_pid_blocks.slx).Required: Never use the PID Controller module in the library.Please use the basic library module to build a parallel form PID: the error e = r-y is a Sum (+ -); the P branch is a Gain; the I branch is a Gain + Integrator; the D branch is a Derivative + Gain; the output of the three branches is another Sum (+ + +) to compose the control amount u.The controlled object uses Transfer Fcn, numerator [1], denominator [1 1].Reference input with Step (0 moment step to 1).The output uses Scope to observe the controlled amount, and briefly describes the names and parameters of each module.

* * Expected effect: * * Get the new model * * `demo_pid_blocks.slx` * * with the structure of * * Step → error and → P/I/D three branch → control and → object → feedback to the error sum * *, no * * PID Controller * * module throughout.

! [insert image description here] (./8.png)!
[Insert image description here] (./9.png)  
 * * Sometimes these modules made by AI are stacked together, and there is no good typography. At this time, you can choose to use Simulink's built-in typography function.  
 You can also rearrange the format with AI to make the model understand better * *

---

# # # # ③ Take a simple closed-loop circuit and simulate the picture

`` `text
Please use Simulink MCP: load the demo_pid_blocks.slx just now.Confirm that there is unit negative feedback in the model (reference, error, controller, object, Scope).Then call simulate: model_name is demo_pid_blocks, stop_time is 20, return_figures is true.After the simulation, the returned graph is displayed, and one or two sentences are used to describe whether the controlled quantity converges and has obvious oscillations from the step response.

• • Expected effect: * * The PNG curve returned by * * simulate appears in the dialog * * (from Scope, etc.); if the error “model not found” is reported, please perform * * ② * * first or change the path to the actual position of your native `.slx’.

• • Tip: * * MATLAB may * * cold-start for 15-20 seconds * * when calling the MCP tool for the first time. If there is no figure in the text only, check whether * * Scope * * is connected to the model and return_figures is * * true * *.

! [insert image description here] (./10.png)! [Insert image description here] (./11.png)

# # X. Reference links

• Warehouse:https://github.com/sohumsuthar/simulink-mcp
• MCP protocol:https://modelcontextprotocol.io
• MathWorks documentation: Search the official website for * * MATLAB Engine for Python * * and the current distribution’s * * Supported Python Versions * *