Simulink Agentic Toolkit

Give your AI coding agent the ability to read, build, edit, and test Simulink® models using Model-Based Design best practices.

---

What It Does

The Simulink Agentic Toolkit packages MathWorks® Model-Based Design expertise for AI coding agents. It connects agents to Simulink through the Model Context Protocol (MCP), giving them both the ability (tools) and the knowledge (skills) to work with Simulink models effectively.

• 6 MCP tools for reading, editing, querying, testing, and checking Simulink models
• 7 agent skills encoding MBD best practices for model building, plant specification, testing, requirements, and tool initialization
• Per-agent manifests for Claude Code, Cursor, Codex, Copilot, Amp, and Gemini CLI — install once, skills and MCP are configured automatically
• Automated setup that configures MATLAB® paths, enables MCP attachment, and validates your installation

---

How It Works

┌───────────┐       ┌───────────┐       ┌──────────┐
│ AI Agent  │◄─MCP─►│MCP Server │◄─────►│ MATLAB / │
│ (Claude,  │       │ (MATLAB   │       │ Simulink │
│  Cursor,  │       │ MCP Core) │       └──────────┘
│  Copilot) │       └───────────┘
└───────────┘
      ▲
      │ reads
┌─────┴─────┐
│  Skills   │
│ (MBD best │
│ practices)│
└───────────┘

Your agent reads skills for domain knowledge, then calls MCP tools to interact with MATLAB and Simulink. The MATLAB MCP Core Server bridges the connection (downloaded during setup).

---

Supported Platforms

Platform	Setup	Notes
Claude Code	Automated	Also supports no-clone marketplace install (skills only)
GitHub Copilot	Automated
OpenAI Codex	Automated
Gemini CLI	Automated
Sourcegraph Amp	Automated
Cursor	Manual	Untested

Automated setup has been verified with basic workflows on each platform except Cursor. The toolkit is under active development — please report issues if you encounter problems.

Quick Start

Full walkthrough: See the Getting Started guide for detailed instructions, platform-specific notes, verification steps, and troubleshooting.

Prerequisites:

• MATLAB R2023a or later with Simulink
• Supported AI coding agent
• Git™

The Simulink Agentic Toolkit helps you install and configure the MATLAB MCP Core Server, or can be configured to use your existing installation.

Full Setup (recommended)

Clone the repository, launch your agent from the toolkit directory, and ask it to set up the toolkit.

git clone https://github.com/matlab/simulink-agentic-toolkit.git
cd simulink-agentic-toolkit

Launch your agent (claude, codex, gemini, etc.) and ask:

Set up the Simulink Agentic Toolkit

Setup looks for your MATLAB installation(s), downloads the MCP server, writes your agent's global configuration, and registers skills. Once complete, start a new session in any project directory — Simulink tools and skills are available everywhere.

Claude Code — no clone required: If you already have the MCP server configured, you can add skills directly without cloning:

claude plugin marketplace add "https://github.com/matlab/simulink-agentic-toolkit"
claude plugin install model-based-design-core@simulink-agentic-toolkit

This installs skills only. Your existing MCP configuration is not modified. See the Getting Started guide for details.

Already Have the MCP Server?

If you installed the MATLAB MCP Core Server yourself, you just need skills. See Adding Skills Only in the Getting Started guide.

MATLAB Setup (all platforms)

The MCP server connects to a running MATLAB session. Open MATLAB and run:

addpath("/path/to/simulink-agentic-toolkit")
satk_initialize

Verify

In MATLAB, open any Simulink model — your own, or a shipped example like f14:

openExample("simulink/AddBlockToModelFromLibraryExample")       % only needed for R2023b+
open_system("f14")

Then ask your agent:

Describe the structure of the currently open model.

---

MCP Tools

Tool	What your agent can do
model_overview	Explore model architecture — see subsystem hierarchy, interfaces, and how major components connect
model_read	Understand model behavior — inspect blocks, algorithmic expressions, signal flow, and parameter values
model_edit	Build and modify models — add blocks, wire signals, create subsystems, and configure parameters
model_test	Verify requirements — run human-readable Gherkin tests with automatic harness generation (requires Simulink Test)
model_query_params	Inspect any parameter — query block settings, signal properties, solver config, and logging flags
model_resolve_params	Get actual values — resolve workspace variables like Kp to their numeric values across all scopes

---

Agent Skills

Skills are organized in the skills catalog. The core skill group includes:

Skill	What it teaches your agent
building-simulink-models	Best practices for structural model changes — adding blocks, wiring, layout
filing-bug-reports	Generate standalone bug reports for reproducing, investigating, and fixing issues
simulink-agentic-toolkit-setup	MATLAB path configuration, MCP tool initialization, tool selection guidance
specifying-mbd-algorithms	Specify algorithms for MBD — system specs, architecture specs, implementation and test plans
specifying-plant-models	How to specify plant models for closed-loop simulation
testing-simulink-models	How to test model behavior — reproduce issues, verify changes, regression tests
generate-requirement-drafts	Requirements generation — prefers Requirements Toolbox (.slreqx) with traceability links when available, falls back to structured YAML

---

Repository Structure

simulink-agentic-toolkit/
├── .claude-plugin/           # Claude Code + Copilot manifest
├── .cursor-plugin/           # Cursor manifest
├── .agents/plugins/          # Codex marketplace registry
├── .codex-plugin/            # Codex plugin definition
├── gemini-extension.json     # Gemini CLI MCP config
├── skills-catalog/           # Agent skills (not auto-discovered)
│   ├── model-based-design-core/  # Core MBD skills (6 skills)
│   └── PRODUCT-GROUP-NAME/      # Placeholder for additional skill groups
├── tools/                    # MCP tool implementations
├── satk_initialize.m         # MATLAB setup entry point
└── research-previews/        # Curated example tasks

---

Research Preview: Agentic Task Explorer

The Agentic Task Explorer provides curated, multi-step tasks that demonstrate what agents can do with Simulink — model understanding, creation, modification, testing, bug fixing, and verification. Each task includes Simulink models and supporting files, ready to go.

slAgenticTaskExplorer

Select a task from the interactive UI. The explorer stages it into an isolated workspace with all required files, then opens your coding agent. Each task presents step-by-step prompts — copy each prompt into your coding agent and watch it work.

This is a research preview. Behavior and interfaces may change.

---

Requirements

• MATLAB R2023a or later with Simulink
• Simulink Test (optional) — required only for model_test
• System Composer (optional) — enables architecture modeling and component analysis
• Simscape (optional) — enables physical modeling domain support
• Stateflow (optional) — enables state machine and chart analysis
• A supported AI coding agent (see Supported Platforms)

---

AI Model Capability Guidance

This toolkit relies on strong multi-step reasoning, tool use, and coding performance from the AI model.

We have tested the toolkit with higher-capability models, including Claude Opus and Sonnet, OpenAI GPT-5 models, and Gemini Pro models, and have generally seen good results on demanding workflows.

Model capability has a significant impact on quality. In our testing, lightweight or lower-capability models were less reliable for tasks such as model construction and complex edits, and were more likely to produce incomplete or incorrect results. These models may still be sufficient for simpler tasks, but for the best overall experience we recommend using a higher-capability model.

---

Documentation

Resource	Description
Getting Started	Setup tutorial with per-agent instructions and troubleshooting
Skills Catalog	Browse all agent skill groups and individual skills

Trademarks

MATLAB and Simulink are registered trademarks of The MathWorks, Inc. See mathworks.com/trademarks for a list of additional trademarks. Other product or brand names may be trademarks or registered trademarks of their respective holders.

Reporting Bugs

If you encounter a bug, use the filing-bug-reports skill to generate a report before opening a GitHub issue. Ask your agent:

File a bug report for this issue

The skill automatically captures environment details, reproduction steps, and error output — producing a complete report in your workspace. Then open a bug report and paste the generated report. Be sure to run the skill in the same session where the bug occurred, since it uses conversation context to reconstruct what happened. If the issue did not occur in a chat session, describe the issue as best you can to the agent, then ask it to file a bug report.

Contributing

We welcome feedback through GitHub Issues. Pull requests are reviewed for ideas and feedback but are not merged from external contributors. See CONTRIBUTING.md for details.

Support

MathWorks encourages you to use this repository and provide feedback. To request technical support or submit an enhancement request, create a GitHub issue or email [email protected].

When using the Simulink Agentic Toolkit and MATLAB MCP Core 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 MATLAB MCP Core server may only be used with MATLAB installations that are used as a Personal Automation Server. Use with a central Automation Server is not allowed. Please contact MathWorks if Automation Server use is required. For more information see the Program Offering Guide (MathWorks).

---

Copyright 2025-2026 The MathWorks, Inc.