SkinTokens for Agents

SkinTokens for Agents

Rigs a 3D mesh with a predicted skeleton and skin weights, through a REST API, MCP tools or a web page, on ZeroGPU hardware. For 3D artists, game developers and AI agents.

Made by townbox on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: SkinTokens for Agents is an auto-rigging service for 3D meshes. You send a mesh in glb, gltf, obj or fbx format, and you get it back rigged, with a predicted skeleton and skin weights for every vertex. It is offered as a REST API, as MCP tools and as a Gradio page, and it runs on ZeroGPU hardware.

Who it suits: It suits 3D artists, game developers and AI agents that need a rig for a mesh without building one by hand.

Industry | Who uses it | How they use it | Example
Game development | Developers | Rig character or creature meshes automatically | Send a quadruped mesh and get a rigged GLB
3D art | Artists | Get a starting skeleton and skin weights | Preview the bones over the mesh
AI agents | Developers | Let an agent call the rigging as a tool | Use the MCP tool named rig

If you already have a finished rig and only want to animate it, you can skip it.

Where to use it:

• A web browser, on the Space's page on Hugging Face
• An MCP client, such as Claude Code, using the service's MCP endpoint
• Any program that can call a REST API with a Hugging Face token
• A local copy of the files, run on Python 3.12.12 under Windows, macOS or Linux

Use case | Who | What they get back
Rig a mesh | Artists and developers | A rigged GLB, rig JSON, weights as NPZ and a preview
Keep an existing skeleton | Developers | Predicted weights for a skeleton that is already there
Preserve the original file | Artists | The rig moved onto the untouched input, so materials and textures survive

How to set it up:

• A web browser, which is all the online page needs
• A Hugging Face account and token, to call the API, since GPU time is billed to the caller
• Python 3.12.12 and the packages in requirements.txt, for a local copy

1. Open the Space's page on Hugging Face to use it online. Nothing needs installing.
2. To call it from code or an agent, send your Hugging Face token as a bearer token with each request.
3. For a local copy, install the Hugging Face hub package and download the Space's files.
4. Move into the downloaded folder.
5. Install Gradio 6.17.3, then the packages from requirements.txt, and start the app with python app.py.

pip install -U huggingface_hub
hf download townbox/skintokens-for-agents --repo-type space --local-dir skintokens-for-agents
cd skintokens-for-agents
pip install gradio==6.17.3
pip install -r requirements.txt
python app.py

How to use it: Send a mesh of up to 60 MB, as a file upload, as a link or by naming a bundled example. Rigging a mesh with about 15k vertices takes roughly 20 seconds in one call, once a worker is warm. The page also has a card for your agent and a card for MCP setup.

Tool | What it does
rig | Rigs a mesh and returns the rig files
list_examples | Lists the bundled examples
get_result | Fetches the result of a rigging job
view_rig | Shows a preview of a rig
service_info | Reports information about the service
guide | Returns the guide for agents

Example: Ask for the bundled giraffe example with a seed of 7. The service returns a rigged GLB, the rig as JSON with joint names and hierarchy, the weights as NPZ, and a preview with the skeleton over the mesh.

For newcomers and experts: Open the Space page, look at the bundled examples and play one in the browser. Each example plays the rig with its bones drawn and a short test animation that stresses the skin weights.

The model is SkinTokens, also called TokenRig. A mesh encoder conditions a Qwen3-0.6B decoder that writes skeleton tokens and then learned skin tokens per joint, which are decoded into skinning weights. Only the model's forward pass runs on the GPU, and a Blender 4.5 worker handles import and export. The keep_skeleton option predicts weights for an existing skeleton, and preserve_original moves the rig onto the untouched file.

You send a mesh
A Blender worker reads and normalizes it
TokenRig predicts skeleton and skin tokens
The tokens decode into skin weights
The rigged GLB and previews are written

Good to know: GPU time is billed to the caller and never to the Space, because the Hub attaches the caller's ZeroGPU token.

• License: MIT, for the code and the weights
• Cost to run: a Hugging Face account token, with the GPU time charged to the caller's ZeroGPU quota
• Data it reaches: the meshes you send, up to 60 MB, and the SkinTokens and Qwen3-0.6B models
• Not for: meshes larger than 60 MB

On Hugging Face, 13 November 2026

• Likes: 4
• Built with: gradio
• License: mit

Links

• Space: https://huggingface.co/spaces/townbox/skintokens-for-agents
• Files: https://huggingface.co/spaces/townbox/skintokens-for-agents/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.