MathWorks launches open-source tools for MATLAB AI agents

MathWorks has introduced the open-source MATLAB MCP Server and MATLAB Agentic Toolkit, which let AI agents execute and refine workflows inside MATLAB.

The software is designed to let agents write MATLAB code, run it in a live session, inspect outputs or errors, and repeat the process until they reach a result. Engineers remain responsible for validating outputs and applying their own judgment.

The launch reflects a broader push by software suppliers to connect generative AI systems more closely with established engineering tools rather than relying on text generation alone. MathWorks is seeking to place agents inside the same environment used for numerical analysis, modelling, and simulation.

Both products are being released as open-source packages, allowing developers and companies to inspect the code, extend it, and use it in their own environments.

The tools use Model Context Protocol, a standard that helps AI agents connect with external software and services. Through that framework, the products can work with agentic AI platforms including Claude Code, GitHub Copilot, OpenAI Codex, and Gemini CLI.

Engineering focus

MathWorks is framing the move around the idea that engineering teams need results that can be executed and checked, not just code suggestions produced by large language models. By allowing an agent to run code directly in MATLAB, it argues that workflows can be grounded in deterministic computation and executable models.

That matters in engineering settings, where teams often need to compare outputs with expected behaviour, test assumptions, and iterate through revisions before accepting a result. The workflow is aimed at MATLAB users, applied AI engineers building agent-driven processes, and platform teams overseeing AI-assisted engineering environments.

Founded in 1984, MathWorks is known for MATLAB and Simulink, both widely used in industry and academia for algorithm development, modelling, and simulation. The company employs more than 6,500 people across 34 offices worldwide.

Industry analysts see the shift as part of a broader change in how AI is being introduced into engineering organisations. Instead of focusing only on code generation, suppliers and users are increasingly looking at how agents can handle repetitive tasks inside controlled workflows while people retain oversight.

One analyst made that point in comments accompanying the launch.

"As organizations adopt agentic AI in Model-Based Design and engineering, the focus is shifting from code generation to reliable execution within established multi-disciplinary toolchains," said Diego Tamburini, AI Practise Director at CIMdata. "Engineers remain responsible for defining problems, validating outcomes, and maintaining oversight, while AI agents increasingly handle iterative and repetitive tasks-augmenting human efficiency and effectiveness. This reinforces the importance of human-in-the-loop workflows, where real execution and validation underpin trust in AI-driven engineering processes."

Open integration

The open-source element may also matter for companies that want tighter control over how AI tools are introduced into technical environments. In many engineering organisations, platform teams need to inspect integrations, set permissions, and decide how external tools connect with internal software.

By making the MCP server and toolkit available for inspection and modification, MathWorks is offering those teams a way to incorporate agent-based workflows without treating the software as a closed black box. It says this should support interoperability across different AI agent frameworks.

For users, the practical promise is that an agent can move beyond producing draft code and instead take actions inside MATLAB itself. That includes running scripts, reviewing output, spotting errors, and revising the workflow through repeated cycles.

MathWorks says the process is intended to keep the engineer in control. The tools are positioned as a way to support human oversight and validation rather than replace them.

Seth DeLand, Generative AI Product Manager at MathWorks, said the value of agentic AI in engineering depends on direct access to the software already used for design and analysis.

"AI agents are most effective in engineering when they can directly interact with the tools used for design, simulation, and analysis," said Seth DeLand, Generative AI Product Manager at MathWorks. "By enabling agents to execute and iterate MATLAB workflows, we're connecting AI-driven iteration to the same computational environment engineers use to develop and validate their work. This allows teams to move from LLM generated code to executable, testable results within a consistent engineering framework."