Connecting an AI agent to CAD software does not guarantee access to the information or operations an engineer needs. Leo AI’s comparison examines Model Context Protocol servers for SolidWorks, Onshape, Fusion, and Creo, emphasizing differences in ownership, maintenance, and scope.
MCP standardizes how AI applications discover and invoke tools. It does not certify their capabilities. Teams must establish who maintains a connection, which functions it exposes, where it runs, and whether existing permissions protect the data involved.
According to the article, SolidWorks relies on community projects rather than an official Dassault Systèmes server. These projects can expose functions such as custom properties, bills of materials, drawing exports, and macros. However, compatibility and reliability vary, leaving users responsible for evaluation and maintenance.
Onshape offers an official FeatureScript MCP server through its Labs program. It helps AI clients generate, execute, and refine code for reusable custom features. Its focus is creating the instructions that produce geometry, rather than providing assembly analysis or unrestricted model editing.
Fusion separates its official capabilities between two servers. Fusion Data MCP handles hubs, projects, folders, permissions, and information. It connects through Autodesk authentication and does not manipulate geometry. Fusion MCP instead connects to a running design session, supporting modeling operations and changes through Fusion’s API and scripting capabilities. Choosing the correct server therefore depends on whether the task concerns project administration or work inside a model.
For Creo, the article describes an experimental MCP interface built around the community CREOSON toolkit. Demonstrated tasks include drawing generation and assembly configuration changes, but incomplete API coverage and unreliable sketch automation remain limitations.
The comparison also promotes Leo’s own search product for information spanning CAD, PDM, PLM, and ERP. That commercial perspective underscores a useful distinction: controlling one CAD application and retrieving engineering knowledge are separate problems requiring different capabilities.
