Manifold3d

English | 简体中文

License: Apache-2.0 Ask DeepWiki

Describe a part. Refine it in 3D. Export STL for your slicer.

Manifold 3D brings AI 3D modeling to GitHub Copilot and Claude Code. Your assistant writes parametric TypeScript, validates the geometry, and opens an interactive preview. Use it for code-driven parametric CAD and 3D printing projects, from phone stands and organizers to keycaps and terrain models.

From prompt to model

Copilot Canvas showing a parametric phone stand with a tilted backrest and widened 22 mm cable slot

Phone stand in Copilot Canvas, with the cable slot widened to 22 mm. Compare the 14 mm and 22 mm slots and reproduce the model.

View the parametric phone stand source.

Ask for a phone stand with a cable slot. In Copilot's native Canvas, select the slot and ask to make it wider without changing the stand's footprint. Review the updated model, then choose Export → Export STL.

The MCP plugin offers the same modeling engine in a browser Viewer, with annotations for feedback. See the full workflow.

Install

You need Node.js 24 or later and one of the hosts below. Plugins bundle the runtime and Viewer; no separate npm install is needed.

Choose your experience	Plugin	Host
Native Canvas with selection-to-chat	manifold-extension	Copilot CLI / Copilot app with Canvas support
MCP tools and browser Viewer	manifold	Copilot CLI or Claude Code

Copilot — native Canvas

copilot plugin marketplace add zhicwan/manifold3d-mcp
copilot plugin install manifold-extension@manifold3d-mcp

For the MCP/browser experience instead, use the same marketplace and install:

copilot plugin install manifold@manifold3d-mcp

Claude Code — MCP/browser (run inside Claude Code)

/plugin marketplace add zhicwan/manifold3d-mcp
/plugin install manifold@manifold3d-mcp

The plugins can be installed separately. Native Canvas requires a compatible Copilot host; it is not part of the MCP plugin. For standalone MCP, updates, and migration, see the usage guide.

Try your first prompt

Use Manifold to design a parametric phone stand in millimeters: 80 mm wide,
85 mm deep, with a 6 mm base, a backrest tilted 20 degrees from vertical,
and a centered 14 mm front cable slot. Keep the key dimensions as named
parameters. Validate the script, preview the model, and inspect a rendered
view before showing me the result.

Then try: “Widen the cable slot to 22 mm; keep everything else unchanged.”

Start with dimensions, fit requirements, and one clear revision at a time. You do not need to write TypeScript yourself, but the model remains code you can inspect and edit.

Examples and questions

• Browse the samples: a cube, revolved vase, gyroid, keycap set, magnetic photo frame, and more.
• Read the usage guide: installation, local scripts, annotations, Viewer shortcuts, and STL export.

Is this the Manifold library?

No. This project adds assistant tools, validation, and a Viewer around manifold-3d, the underlying geometry library.

Is it a traditional CAD editor or a separate AI model?

Neither. Your existing assistant generates code; the Viewer lets you inspect geometry and point out changes. It is not a sketch-and-constraint CAD editor.

What can I export, and is it ready to print?

The Viewer exports STL, not STEP or 3MF. Validation checks the script and geometry, not manufacturing or use safety. Check dimensions, clearances, wall thickness, materials, and slicer settings before printing.

Contributing

See CONTRIBUTING.md for setup and checks, AGENTS.md for repository guidance, and the Extension README for technical Canvas integration.

License and upstream

Apache License 2.0 — Copyright 2026 Zhicheng Wang.

See NOTICE for upstream attribution.

This project uses and adapts portions of elalish/manifold (Apache-2.0):

• packages/modeling/src/sandbox/garbage-collector.ts
• Documentation under skills/shared/references/