OpenSCAD MCP Server

What is OpenSCAD MCP Server?

A Model Context Protocol (MCP) server that generates parametric 3D models from text descriptions or images using AI image generation, multi-view reconstruction via CUDA Multi-View Stereo (MVS), and OpenSCAD integration. It is designed for makers, designers, and hobbyists who want to create printable 3D models from high-level prompts.

How to use OpenSCAD MCP Server?

Install dependencies (OpenSCAD, CUDA Multi-View Stereo, Python packages), set API keys in a .env file (GEMINI_API_KEY, optional VENICE_API_KEY), then start the server with python src/main.py. Interact via MCP tools such as create_3d_model_from_text, generate_image_gemini, or use the web interface at http://localhost:8000/ui/.

Key features of OpenSCAD MCP Server

• AI image generation via Gemini or Venice.ai APIs
• Multi-view image generation for 3D reconstruction
• Image approval workflow before reconstruction
• 3D reconstruction using CUDA Multi-View Stereo
• Remote processing on LAN servers for heavy tasks
• Parametric export formats (CSG, AMF, 3MF, SCAD)
• Optional 3D printer discovery and direct printing

Use cases of OpenSCAD MCP Server

• Generate a low-poly 3D model from a text description like “a low-poly rabbit”
• Create multiple views of an object from a single prompt and reconstruct a 3D mesh
• Offload CUDA MVS reconstruction to a powerful remote server on the same LAN
• Export a parametric model to OpenSCAD source for further manual refinement
• Discover and send a model directly to a network 3D printer

FAQ from OpenSCAD MCP Server

What image generation options are available?

The server supports Google Gemini API (default) and optionally Venice.ai; you can also upload your own images to skip AI generation.

How does the multi-view workflow work?

Generate multiple consistent views of an object, review and approve each image, then reconstruct a 3D model from the approved set using CUDA MVS.

Can I process reconstruction on a different machine?

Yes: run the remote CUDA MVS server (python src/main_remote.py) on a LAN machine with a CUDA GPU, then configure the main server to discover or specify that remote server.

What export formats are supported?

OBJ, STL, PLY, SCAD, CSG, AM