Skip to content

Latest commit

 

History

16 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

motionmcp-kimodo

License Python Status

Self-hosted MMCP server for Kimodo motion generation.

Run Kimodo on your own GPU and expose it over HTTP so any MMCP client can generate animation. The reference client is Proscenium for Blender — text prompts, paths, and pose constraints on your armature.

Who this is for

This repo is for developers who are comfortable with Python environments, CUDA, and self-hosting ML services. The install assumes you can debug pip, virtualenvs, CUDA drivers, and Hugging Face authentication on your own. See Installation and MotionCorrection for what that involves.

If you are an animator or artist and want Kimodo in Blender without that setup, use the hosted product at animatica.ai — same model family, no local server install. Point the Proscenium plugin at your Animatica endpoint and generate from there.

Alpha — APIs and packaging may change. Tracks motionmcp 0.1.x and Kimodo 1.x.

Features

  • MMCP-native — implements motionmcp.Backbone; capabilities, /generate, glTF responses
  • Kimodo SOMA models — loads Kimodo checkpoints; maps MMCP requests to Kimodo inference
  • Constraint-aware — root paths, effector targets, pose keyframes (see MMCP concepts)
  • Simple CLImotionmcp-kimodo --port 8000 for local or LAN use
  • Embeddable — mount KimodoBackbone in your own FastAPI / ASGI app (Development)

Requirements

Python 3.10+
GPU NVIDIA GPU strongly recommended (CUDA-capable PyTorch)
Build tools CMake + C++ compiler — Kimodo builds MotionCorrection during install (guide)
Git Required for pip install (Kimodo is not on PyPI)

Full setup (Blender addon, troubleshooting): Installation.

Quick start (self-hosted)

pip install "motionmcp-kimodo @ git+https://github.com/animatica-ai/motionmcp-kimodo.git"
motionmcp-kimodo --port 8000

Install Proscenium, set Server to http://localhost:8000, and connect from the N-panel.

Install can take 30+ minutes on first run (PyTorch, Kimodo, MotionCorrection compile). Use pip install -v ... to see progress.

Documentation

Guide Description
Installation End-to-end setup with Proscenium
MotionCorrection C++ build step (common install blocker)
Usage CLI, ports, models, environment variables
Development Architecture and programmatic use
Docs index All guides and external links

Related projects

Project Role
MMCP Motion Model Communication Protocol
motionmcp Python SDK (Backbone, server helpers)
Kimodo Motion diffusion model (dependency)
Proscenium Official Blender client
Implementations All official servers & clients

Contributing

Contributions are welcome.

  1. Check open issues or open one to discuss larger changes.
  2. Clone, install dev deps, and run tests — see Development.
  3. Open a pull request with a clear description and test plan.

Bug reports for install failures: include OS, Python version, and the full pip install -v log (especially the MotionCorrection build).

Community

Animatica AI Discord — questions, install help, Proscenium, and MMCP/Kimodo discussion.

License

Apache License 2.0. Kimodo model weights and third-party deps have separate licenses — see Kimodo and Hugging Face model cards.

About

Apache-2.0 reference backbone wrapping NVIDIA's Kimodo motion model behind MMCP. Run Kimodo on your own GPU.

Topics

Resources

Stars

16 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages