RSS
Tools

Onshape's FeatureScript MCP server turns Claude Code prompts into reusable CAD features

Onshape Labs' FeatureScript MCP server lets Claude Code and other AI clients write, run and fix custom CAD features from plain-language requests. Onshape's examples range from a sprocket generator to a water park design system. Setup steps, use cases and limits.

a water slide in the middle of a swimming pool
Photo by Wasif Mujahid on Unsplash

Updated November 20, 2026: Rewritten and expanded with setup steps and use cases. We removed part-count, prompt-count and timing figures we could not find in Onshape's own pages.

Onshape, the cloud-native CAD and data management platform from PTC, has a new way to work with AI. Instead of asking a model to produce geometry directly, you ask it to write FeatureScript, Onshape's built-in language for CAD features. The FeatureScript MCP Server, an Onshape Labs app, connects AI clients such as Claude, ChatGPT and Gemini to that language. MCP, the Model Context Protocol, is an open standard that lets AI applications interact with external systems. Onshape calls the result text-to-code-to-CAD.

When we first covered this, we gave a part count, a prompt count and a time for Onshape's water park demo. Those figures do not appear in the text of Onshape's blog posts or setup guide; the water park is shown in an embedded video we could not read, so we have taken the numbers out. The water park is still one of Onshape's showcase examples, and it is described below.

How it works

FeatureScript is the same language Onshape's own developers use to build the platform's native tools. A custom feature written in it behaves like any built-in command. It is fully parametric, sits in the toolbar, and works across browsers, iOS and Android. Through the MCP server, the AI can generate FeatureScript, insert it into an Onshape document, run it, read the results and errors, and revise the code until the feature works.

The output is a tool, not a one-off model. A sprocket feature, for example, can generate countless sprocket variations from its parameters. And once the code exists, Onshape points out, it runs like any other custom feature. The AI did the work during development, so later uses spend no AI tokens. If you would rather write the code yourself, Onshape also offers FeatureScript Co-Complete, which helps with syntax and boilerplate while you author the logic; the MCP server is for handing the whole build to the AI.

What teams are building with it

  • Standards-driven parts: a garden hose thread built to ASME V1.20.7, or a roller chain sprocket generator with configurable strand counts and manufacturing annotations. Another example turns a PDF of a flanged bushing into a configurable feature with its size ranges.
  • Reusable libraries: colour maps, material lookups and geometry helpers that later features call by name.
  • Refactoring old models: a piston model with 60 features consolidated into one parametric feature.
  • Analysis tools inside CAD: an FEA beam analysis tool, and a heat exchanger design system following ASME standards that calculates thermal performance.
  • Design systems: a lead screw generator for ACME, buttress and knuckle threads, and wire-shelf tools that produce cut lists and bend data.

The water park is Onshape's example of a whole industry-specific application. Site planning, ride specifications, structural systems, rider behaviour and performance calculations are joined up as a connected set of custom features. The point is that one feature's output drives the next, so the design intent stays traceable from layout to detail.

How to set it up with Claude Code

Onshape's tech tip gives the steps for Windows and macOS. First, sign in to Onshape, open the Onshape App Store, find FeatureScript MCP Server and click Subscribe. Then go to your Application Settings and confirm Onshape appears in the list. On Windows only, add C:\Users\USERNAME\.local\bin to your system PATH. If you don't have Claude Code, the guide installs it with curl -fsSL https://claude.ai/install.sh | bash on a Mac or irm https://claude.ai/install.ps1 | iex in PowerShell, and claude --version confirms it worked. Then add the server:

claude mcp add --transport http onshape https://fs-mcp.labs.onshape.app/mcp

Type claude to launch Claude Code and follow the prompts until it asks you to authorize your subscription; click Authorize in the browser window that opens, close it, and press Return in the terminal. Next, type /mcp inside Claude Code and follow the prompts to connect your Onshape account, then press Escape. To test it, create a new Onshape document, copy its URL and type: Create a new Feature Studio named FS1 in this document: [URL]. FS1 should appear in the document. If Claude later says the tools are unavailable, re-authenticate with /mcp. If your company has its own Onshape API allocation, tell Claude: Use the Onshape company API allocation, not my user API allocation.

Try it with your AI. A first request to try once the server is connected (replace the URL). It creates a new Feature Studio and code in that document, so use a test document:

In this Onshape document: [URL], create a Feature Studio and write a custom feature that
generates a flanged bushing. Expose inner diameter, outer diameter, flange diameter,
flange thickness and length as parameters with sensible limits, validate that the
diameters are consistent, and add the feature to the document's toolbar. Test it with
three different sizes and fix any errors before you finish.

Limits

The server is an Onshape Labs release, which Onshape describes as an early-access program for emerging technologies and experimental workflows, so commands and screens may change; check Onshape's current instructions if a step looks different. You need an Onshape account. Onshape says the AI identifies errors and revises the code until the feature works, and that the complexity of features it can generate will grow as models improve. Our own advice, not Onshape's: have an engineer check the FeatureScript and its output against the standard before using it for production parts.

Our take

Writing code that creates geometry, rather than producing geometry directly, is the right design for engineering work. You get something deterministic, editable and shareable, and you pay for the AI once instead of every time a part is regenerated. For teams that already keep their standards in PDFs and spreadsheets, turning those into custom features is the obvious first project. It is also a useful lesson for other tool builders: give the model a language and a test loop, not a single-shot output.

Source: 5 Ways Teams Can Leverage the Onshape FeatureScript MCP Server (Onshape), FeatureScript MCP Server Enables Text-to-Code-to-CAD (Onshape), Tech Tip: How to Connect the FeatureScript MCP Server and Claude Code (Onshape)

This story first appeared in the claudefox daily briefing for November 6, 2026.