{"id":19058,"date":"2026-01-26T15:00:26","date_gmt":"2026-01-26T20:00:26","guid":{"rendered":"https:\/\/blogs.mathworks.com\/deep-learning\/?p=19058"},"modified":"2026-01-29T17:13:28","modified_gmt":"2026-01-29T22:13:28","slug":"optimization-agent","status":"publish","type":"post","link":"https:\/\/blogs.mathworks.com\/deep-learning\/2026\/01\/26\/optimization-agent\/","title":{"rendered":"Building an Optimization Agent with MATLAB MCP Core Server"},"content":{"rendered":"
\r\n
\r\n
<\/h6>\r\n\r\n\r\n\r\n
\"\"<\/td>\r\nCo-author: Tom Couture<\/a><\/strong>\r\n
<\/h6>\r\nTom Couture is the Product Manager for Optimization. In this blog post, he joins me to demonstrates how to use the new MATLAB MCP Core Server to build optimization agents.<\/td>\r\n<\/tr>\r\n<\/tbody>\r\n<\/table>\r\n
If you are like Tom, you have probably spent hours setting up optimization problems in MATLAB\u2014defining objective functions, constraints, choosing solvers, tweaking options. Now, what if I told you that you can have a conversation with an AI agent that does all of this for you, like a mini Tom in your pocket, while <\/span>actually running<\/span> the optimization in your local MATLAB?<\/span><\/h6>\r\n<\/div>\r\n
That is where the MATLAB MCP Core Server<\/span> comes in, combined with the powerful Optimization Toolbox\u2122<\/span>. In this post, we are going to walk you through how to create an optimization agent that can solve real engineering problems\u2014from simple constrained optimization to computationally expensive CFD-based design optimization using surrogate models.<\/div>\r\n
<\/h6>\r\n

The First Thing Your Agent Should Do: Check the Toolboxes<\/h2>\r\n
One of the first features of the MATLAB MCP Core Server I demo is the detect_matlab_toolboxes<\/span> tool. Before your agent starts writing optimization code, it should verify that you have the necessary toolboxes installed.<\/div>\r\n<\/div>\r\n
<\/h6>\r\n\"\"\r\n
<\/div>\r\n
<\/h6>\r\n
\r\n
Here is a sample prompt to your AI assistant (Claude Desktop\u00ae, GitHub Copilot\u00ae, or any MCP-compatible client):<\/div>\r\n
\"Check which MATLAB toolboxes I have installed, specifically looking for Optimization Toolbox and Global Optimization Toolbox.\"<\/span><\/div>\r\n
The agent will call the detect_matlab_toolboxes<\/span> tool and return something like:<\/div>\r\n
\r\n
Optimization Toolbox - Version 25.2 (R2025b) \u2713<\/span><\/div>\r\n
Global Optimization Toolbox - Version 25.2 (R2025b) \u2713<\/span><\/div>\r\n<\/div>\r\n
This is crucial because the solver you can use depends entirely on what you have licensed. No Optimization Toolbox? The agent will know to fall back to fminsearch<\/span>. Have Global Optimization Toolbox? Great, surrogateopt<\/span> is on the table for those expensive simulations.<\/div>\r\n<\/div>\r\n
<\/h6>\r\n
<\/h6>\r\n
\r\n

Example 1: Structural Beam Design Optimization<\/h2>\r\n
Let me demo to you a classic engineering optimization problem: designing a cantilever beam for minimum weight while satisfying stress and deflection constraints.<\/div>\r\n
<\/h6>\r\n

The Problem<\/h3>\r\n
We want to minimize the weight of a rectangular cross-section beam subject to:<\/div>\r\n