Home 9 AI 9 MCP Opens SolidWorks to AI Agents, but Engineering Judgment Remains Human

MCP Opens SolidWorks to AI Agents, but Engineering Judgment Remains Human

by | Sep 21, 2026

Leo explains what AI agents can control through SolidWorks MCP servers today and why software access alone does not provide the engineering intelligence needed for autonomous CAD.
Source: Leo AI.

 

Model Context Protocol, or MCP, is emerging as a standard way for AI assistants to interact with external software and data. For SolidWorks, however, Leo argues that MCP should be viewed as a connection mechanism rather than a new form of CAD intelligence.

MCP provides a standardized format through which an AI application can discover and call tools. In SolidWorks, community-developed MCP servers connect AI agents to the software’s long-established COM automation API, the same interface used by VBA macros and many third-party add-ins. MCP therefore makes existing capabilities easier for AI agents to access rather than adding fundamentally new CAD functions.

Current implementations can read and modify custom properties, retrieve mass properties and bill-of-materials data, run existing macros, export STEP, PDF, and DXF files, and create basic sketches and features. One community project exposes more than 100 tools, but that range largely reflects decades of functionality already available through the SolidWorks API.

Important limitations remain. Dassault Systèmes does not currently provide an official SolidWorks MCP server, and available projects are community-built. More importantly, MCP cannot reliably interpret an open-ended engineering instruction and decide what geometry, material, tolerance, or feature sequence is technically appropriate.

Deployment also raises security concerns. SolidWorks automation runs locally under the logged-in Windows user’s permissions. An AI agent could therefore inherit access to CAD files, network resources, and PDM privileges. Leo recommends starting with read-only tools and testing automation on copies of production files.

The article ultimately separates software control from engineering intelligence. MCP can deliver commands to SolidWorks, but another layer must determine which commands make sense. For reliable AI-assisted CAD, engineering teams still need technical knowledge, company standards, design history, controlled permissions, and validation alongside the connection itself.