Nova3D MCP
3D generation MCP server: from a description the agent gets a GLB model with named parts, joints and a Blender script
- 755 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The server sends descriptions and reference images to the paid Nova3D service and spends account credits
- Sign-in happens in the browser and the session is stored locally while it works
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/nova3dDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The server sends descriptions and reference images to the paid Nova3D service and spends account credits
- Sign-in happens in the browser and the session is stored locally while it works
This is third-party code. Review the repository files before installing.
What it does
Nova3D builds 3D objects and scenes as editable programs: parameters, parts and motion are written in code that can be inspected and rebuilt. The `nova3d-mcp` MCP server opens that pipeline to agents. A model writes construction code for Blender, Blender runs and validates it, and the output is a GLB where parts have names and a hierarchy and joints are real. The main tool `generate_3d` takes a description and, optionally, a reference image in base64. The reply carries a GLB link, a link to the editing session in the Nova3D app, a list of parts, the joint count and a code artifact. The artifact is passed to `regenerate_part`, `add_part` and `articulate_model` to change a single part without rebuilding the whole object. `nova3d_login` handles the first sign-in and `nova3d_status` shows readiness and remaining credits.
Who it is for. Developers of games, configurators and robotics simulations who need 3D assets with separate parts instead of a fused mesh.
Good fit when
- A configurator showcase needs a vending machine with a separate door, glass and buttons so each part can change
- A robotics simulation needs an object with joints rather than a single mesh
- A finished model is edited surgically: regenerate one part and add a new one from a description
- A Blender script is needed as a base that is refined by hand
Not a fit when
- You need a local generator with no account: generation goes through the paid Nova3D service and requires sign-in and credits
- You want to view the model inside the editor: the preview opens in the browser through the session link
- You need your own model provider key: the server does not offer such a mode
Example request
Generate a vending machine with a separate door, glass panel, coin slot, button grid and interior shelvesMCP
- Transport
- stdio
- Authentication
- OAuth
Environment variables
NOVA3D_TOKENsecret- Fallback API key from app.nova3d.xyz/api-key when browser sign-in is unavailable
NOVA3D_APP_URL- App URL used for session links, for example https://app.nova3d.xyz
Limitations
Initial generation runs through the paid GraphFlow v2 workflow, so after installation you need a browser sign-in with `nova3d_login` and credits, which `nova3d_status` reports. The server has no mode with your own provider keys. The preview opens only in the browser through `conversation_url`, and there is no embedded 3D view in the editor. The repository holds the description, datasets and a benchmark, while the server code is published as the `nova3d-mcp` package on PyPI. Running it needs `uv`. The model is picked by a parameter: gemini by default, claude-sonnet, claude-opus or gpt-5.5.
How to disable. Remove the server from the client (`claude mcp remove nova3d`). Sessions and models stay in your Nova3D account and are deleted there.
FAQ
How do I connect Nova3D to Claude Code?
Run `claude mcp add nova3d -- uvx nova3d-mcp`, then call `nova3d_login` and finish the sign-in in the browser tab. Installing the server and signing in to Nova3D are separate steps.
What does generate_3d return?
A `glb_url`, a `conversation_url` for the session in the app, a `parts` list, a `joint_count` and the code and model artifacts that feed later edits.
Are credits required?
Yes. Initial generation runs through a paid workflow, and if `nova3d_status` returns `next_action` equal to `purchase_credits` you follow the purchase link.
Can I set a token by hand?
Only as a fallback when browser sign-in is unavailable: the `NOVA3D_TOKEN` variable with a key from app.nova3d.xyz/api-key.
Which clients are supported?
Codex, Cursor, VS Code, Visual Studio and Claude Code. Each has a command or config file using `uvx nova3d-mcp`.
Discussion
No questions or reviews yet.
Related
MCP serversA remote MCP server with 40 tools: a 3D model from text or a photo, rigging, animation, retexturing and GLB checks
Blender MCP
MCP for Blender
An MCP server and Blender addon: the agent creates and edits objects, materials and scenes, pulls assets and runs Python in Blender
A local MCP server and CLI: the agent writes JavaScript for a 3D model, checks it on renders and exports a GLB with source
An MCP server for generating and editing images with Gemini: picks the model per request, renders up to 4K in any aspect ratio
MCP server from the Vectreal CLI: the agent uploads a GLB, tunes scene lighting and cameras, publishes it and returns embed code
ldraw-nova gives an AI agent tools to build a LEGO model in the LDraw format from a prompt, with real parts and collision checks
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Nova3D (this card) | yes | OAuth | 755 | |
| blenderwright | no | not required | 196 | drives Blender directly from agent commands, with no generation of parts from a description |
| build123d-mcp | no | not required | 151 | a CAD part built step by step in build123d code rather than generated whole from a description |
README in short
The main text presents Nova3D as a workspace for programmable 3D: nine capabilities from executable source to export into CAD and simulators, a separate robotics block and a description of datasets. Connecting to agents is covered in the package documentation on PyPI.