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.

• 7 MCP tools for reading, editing, querying, testing, and checking Simulink models
• 9 agent skills encoding MBD best practices for model building, simulation, plant specification, testing, requirements, project management, and architecture modeling
• Automated setup via a MATLAB® function that installs the MCP server, configures your agent, and registers skills
• Supports Claude Code, Copilot, Codex, Amp, and Gemini CLI

How It Works

┌───────────┐       ┌───────────┐       ┌──────────┐
│ AI Agent  │◄─MCP─►│MCP Server │◄─────►│ MATLAB / │
│ (Claude,  │       │ (MATLAB   │       │ Simulink │
│  Codex,   │       │ 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

Automated setup has been verified with basic workflows on each platform. 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

Automated Setup (Recommended)

The setupAgenticToolkit function handles installation, configuration, updates, and uninstallation for both the MATLAB and Simulink Agentic Toolkits. Download agenticToolkitInstaller.mltbx from the latest release, install it in MATLAB, then run:

setupAgenticToolkit("install")

This downloads the MCP server binary and toolkit files to ~/.matlab/agentic-toolkits/, then walks you through configuring your first coding agent (MCP server entry + skill registration). To set up additional agents later, run setupAgenticToolkit("configure"). To update to the latest version, run setupAgenticToolkit("update"). If your organization uses a CLI wrapper, pass AgentCLI="claude-code=/path/to/wrapper" during configure.

Existing users: If you previously set up the toolkit using the agent-driven workflow, you must uninstall that setup first. See Migrating from a Previous Installation in the Getting Started guide.

Manual Setup

If you already have the MATLAB MCP Core Server installed or prefer full control, you can configure the toolkit manually. See Manual Setup in the Getting Started guide.

MATLAB Setup (all platforms)

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

addpath("~/.matlab/agentic-toolkits/simulink")
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

Agent Skills

Skills are organized in the skills catalog by product area.

Model-Based Design Core — core MBD skills for building, testing, and specifying Simulink models:

Model-Based System Engineering — MBSE skills for System Composer architecture models:

Repository Structure

simulink-agentic-toolkit/
├── skills-catalog/           # Agent skills (not auto-discovered)
│   ├── model-based-design-core/          # Core MBD skills (8 skills)
│   └── model-based-system-engineering/   # MBSE skills (1 skill)
├── tools/                    # MCP tool implementations
├── satk_initialize.m         # MATLAB session 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; required for building-architecture-models skill
• Requirements Toolbox (optional) — enables requirements traceability; used in building-architecture-models skill
• 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

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 [