trellis_mcp

trellis_mcp

2
entertainment_and_media

Trellis MCP Server bridges AI assistants and Trellis using the Model Context Protocol, facilitating the generation of 3D assets from natural language. It is open-source and supports local deployment, emphasizing fast and memory-efficient operations.

Trellis MCP Server

Trellis MCP provides an interface between AI assistants and Trellis via Model Context Protocol (MCP).

Disclaimer

This project shows a very minimal integration of MCP with Trellis: a lightweight and opensource text-to-3d/image-to-3d 3DAIGC model. Compared with existing rodin integration in blender-mcp and tripo integration, it has following advantages:

  • Faster and memory-efficient: You can deploy TRELLIS locally with only 8GPU+ VRAM, while can generate a textured mesh from text in only ~15s(10s with more vram).
  • FREE: You DON'T have to pay expensive API from Rodin/Meshy/Tripo.

BUT IT HAS FOLLOWING LIMITATIONS:

  • Trellis is open-source and there is no off-the-shelf API model providers, you have to deploy it by yourself (refer to README).
  • The API/Prompt has NOT been fully tested/tuned, may suffer from stability issues.

So use it at your own risk.

Demo

A minimal demo for generating a single object, more complicated prompt with blender-mcp is under tuning.

Demo

Features

  • Generate 3D asset from natural language(TEXT) using Trellis API and import into blender
  • Generate texture/materials from natural language(TEXT) for a given 3D mesh using Trellis API and import into blender

Roadmap

Prerequisites

Installation

1. Trellis blender addon
  1. Download Trellis Blender Addon from here
  2. Open Blender -> Edit -> Preferences -> Add-ons -> Install from file -> Select the downloaded addon -> Install
  3. In 3D Viewport -> View3D > Sidebar > TRELLIS -> Start MCP Server
2. Configure API backend

As trellis is a free open-source text-to-3d model, you need to deploy your own trellis API backend with reference to README

# clone an API fork of trellis 
git clone https://github.com/FishWoWater/TRELLIS
cd TRELLIS

# EDIT BACKEND URL in trellis_api/config.py

# configure the # of text workers and start ai worker
# no need for image workers 
python trellis_api/ai_worker.py --text-workers-per-gpu 1 --image-workers-per-gpu 0

# start web server 
python trellis_api/web_server.py 

# or on windows local dev 
python trellis_api/web_server_single.py
3. Configure the MCP server on Windsurf/Cursor/Claude
{
    "mcpServers": {
        "trellis-blender": {
            "command": "uvx",
            "args": [
                "trellis-mcp"
            ]
        }
    }
}

Acknowledgements