PTC has launched the FeatureScript MCP Server for its Onshape CAD and product data management platform, allowing engineers to create reusable parametric CAD features from natural-language instructions rather than writing the underlying FeatureScript code manually.
The capability is available through Onshape Labs and connects compatible AI clients to Onshape through Model Context Protocol, or MCP. PTC lists Claude, ChatGPT, and Gemini among the coding models that can interact with FeatureScript through the server.
An engineer describes the function required, after which the AI can generate FeatureScript, insert the code into an Onshape document, run it, assess the result, identify errors, and revise the implementation. The process can continue through building, testing, debugging, and refinement until the custom feature behaves as intended.
FeatureScript is Onshape’s built-in programming language for creating and modifying CAD features. It is the same language used for native Onshape tools, rather than a separate macro system layered over the modelling environment, meaning completed custom features remain parametric and behave like other commands in the platform.
The MCP Server changes the route into that environment rather than replacing it. FeatureScript has long allowed companies to automate repetitive geometry, calculations, validation checks, and product-specific design procedures, but creating those tools normally required somebody able to translate the engineering logic into code.
Natural-language generation lowers that programming threshold. A designer who understands how a standard hydraulic port, fastening detail, manufacturing allowance, or product-specific feature should behave can describe the requirement and use AI to create the initial implementation without first learning FeatureScript syntax.
The result is different from generating a one-off model from a prompt. A completed FeatureScript feature can be reused across designs, shared with other engineers, edited later, and incorporated into a normal parametric modelling workflow. PTC describes the approach as text-to-code-to-CAD rather than simply text-to-CAD.
That distinction gives the technology a more conventional engineering-automation role than much of the generative AI currently being attached to design software. Engineering organisations already encode internal knowledge in templates, spreadsheets, scripts, checklists, and proprietary CAD features. The MCP Server provides another route for turning those rules into reusable software.
A company could, for example, create a custom tool around an established component family or manufacturing standard, then distribute that feature to designers who no longer need to recreate the same modelling sequence manually. Once the code exists, Onshape says it can continue to run like any other custom feature without returning to an AI model each time it is used.
That persistence also makes validation more significant. AI can reduce the effort required to write a feature, but it does not establish that the underlying engineering requirement is correct. A tool can generate perfectly valid geometry while still embodying the wrong clearance, tolerance, material allowance, design rule, or manufacturing assumption.
Reusable automation magnifies that risk because a flawed feature can be used repeatedly across several projects. Engineering teams will therefore still need ownership, review, testing, documentation, and version control around custom tools, particularly where they capture company standards or influence released product geometry.
The launch materially advances Onshape Labs functionality outlined earlier this summer. FeatureScript MCP Server was presented in July as part of the experimental AI programme; PTC’s 13 August announcement confirms that the capability is now available through Onshape Labs.
Onshape’s own technical material describes Labs as an early-access environment for emerging technologies and experimental workflows. That status gives engineering teams another reason to separate experimentation from unrestricted production use until internally generated tools have been reviewed against their own release and quality procedures.
The server also sits alongside FeatureScript Co-Complete, a separate AI-assisted capability that provides context-aware coding suggestions while the engineer remains directly responsible for authoring the logic. The two approaches offer different degrees of automation: one generates and iterates code from a described outcome, while the other assists someone already working inside FeatureScript.
PTC’s launch therefore makes custom CAD programming accessible to a wider group of users without removing the engineering work around the resulting tool. Engineers can spend less time translating a known process into syntax, but the design rules, test cases, and acceptance criteria still have to come from somewhere more dependable than a prompt.
FeatureScript MCP Server is available now through Onshape Labs. The next useful measure will be how engineering teams govern what it produces — because making internal CAD automation easier to create also makes it easier to distribute a bad rule at considerably greater speed.
