After the MATLAB Agentic Toolkit debuted final week, we’re very completely happy to launch the Simulink Agentic Toolkit on GitHub at the moment.
Setup
You will note that it is so simple as cloning the repository, launching your AI coding agent ((Claude Code, GitHub® Copilot, Cursor, OpenAI® Codex, Sourcegraph Amp, Gemini
CLI) within the repo, and ask “Arrange the Simulink Agentic Toolkit“.
CLI) within the repo, and ask “Arrange the Simulink Agentic Toolkit“. In MATLAB, set up the toolkit
matlab.addons.set up(“~/.native/share/MATLABMCPCoreServerToolkit.mltbx”)
Then share the session to allow the AI agent to hook up with it:
addpath(“/path/to/simulink-agentic-toolkit”)
satk_initialize
What’s within the Toolkit?
Along with the 5 instruments accessible within the MATLAB MCP server, the toolkit provides 7 Simulink-specific instruments:
- model_overview
- model_read
- model_edit
- model_check
- model_query_params
- model_resolve_params
- model_test
These instruments are utilized by 6 expertise:
- building-simulink-models: Finest practices for mannequin building and adjustments
- specifying-mbd-algorithms: Specify algorithms for model-based design — system specs, structure specs, implementation and check plans
- specifying-plant-models: Plant mannequin specification workflow with 4 templates
- testing-simulink-models: Check authoring and validation
- generate-requirement-drafts: Necessities technology
Along with these, the toolkit has a filing-bug-reports ability. In the event you run into points, ask one thing like “Use the filing-bug-reports to report all the issues encountered throughout this session“. This may generate one markdown file per subject. Ship that to technical assist!
My First Instance
I requested the agent:
Make me a plan to create a simulation of the thermal conduct of typical disk brakes in a median sedan automobile. The thermal mannequin ought to embrace related warmth switch mechanisms between the pad, disk and setting. Embrace plans to create checks to train the mannequin for typical driving maneuvers.
One thing like that:
Planning
When asking to implement a mannequin like that, the specifying-plant-models generates markdown information you can assessment and iterate on. Here’s what I acquired after 2-3 iterations:
The plan relies on a template and comprises commonplace sections like:
- Govt abstract
- Drawback assertion
- Objectives & Success Metrics
- Assumptions
- Interface
- References
Then it goes into extra particulars of the implementation.
The plan sometimes contains an ASCII diagram of what the system will appear like. Listed here are 3 iterations I went by the place I requested to incorporate a easy longitudinal automobile with the brake mannequin:
It plans how the equations within the mannequin will likely be componentized in subsystems and referenced fashions:
It described the equations to be carried out in every part:
and the record of parameters:
Implementation
As soon as I used to be proud of the plan, it generated the mannequin and was profitable on the first try. Here’s what it appears like:
By way of parameters, by default it created a MATLAB script defining the variables, however you may ask it to make use of a Knowledge Dictionary as a substitute when you want.
Testing
Testing might be my favourite characteristic of the toolkit.
The check plan listed checks for particular person elements.
It additionally proposed a number of eventualities for the complete closed-loop simulation.
When time got here to implementation, I had the choice to get the check carried out in a MATLAB check class, or within the Simulink Check Supervisor. I like having the checks within the Simulink Check Supervisor, it makes it simple to assessment the outcomes.
Ideas and Methods
Further Expertise
Some expertise like simulink-simulation have been eliminated as a result of they’re included with the toolkit. The remaining expertise just like the profiling ones mustn’t intervene with the toolkit and supply further performance on high of the toolkit.
In the event you create your personal Simulink-related expertise, ask your AI agent to make use of its expertise creator ability to audit your expertise and guarantee they don’t intervene with expertise from the toolkit. When the agent will get conflicting data from totally different expertise, it will get confused and makes extra errors.
Monitoring Instruments Calls
Relying on the AI agent you might be utilizing, there’s a approach to see the main points of the instruments calls it is doing. If all the pieces works advantageous, you must see calls to the instruments listed above: mcp__simulink__model_read, mcp__simulink__model_edit, and many others. In Amp, it appears like this:
In some instances, particularly if the context window turns into lengthy and the agent begins making errors, I’ve seen the agent falling again to strive modifying Simulink fashions utilizing evaluate_matlab_code from the MATLAB Core MCP Server. That is much less environment friendly and has much less consciousness of Simulink, so I like to recommend interrupting the agent and ask it explicitly to make use of the instruments from the Simulink Agentic Toolkit. This often helps getting again on observe.
What’s Subsequent
In the event you’re utilizing an AI agent to work together with Simulink, you actually wish to use the Simulink Agentic Toolkit. The steerage and instruments that the toolkit offers to the AI agent make an enormous distinction.
Give this a try to tell us within the feedback about your expertise.
{ let css=”._richTextNode {white-space: break-spaces;}”;
let head = doc.head || doc.getElementsByTagName(‘head’)[0];
let model = doc.createElement(‘model’);
head.appendChild(model);
model.kind=”textual content/css”;
if (model.styleSheet) {
model.styleSheet.cssText = css;
} else {
model.appendChild(doc.createTextNode(css));
}
}
