Asset MCP - AI Agent Asset Generation Framework

A universal abstraction layer that empowers AI development tools to autonomously generate 2D, 3D, and audio assets through a unified interface, filling the gap in agentic workflows where no existing solution provides comprehensive asset generation capabilities.

Overview

Asset MCP serves as the missing link in AI agent workflows, providing a standardized interface that enables AI development tools to autonomously create a wide range of asset types without human intervention. This abstraction layer bridges the gap between AI agents and various asset generation backends, from commercial APIs to self-hosted solutions.

Purpose

This framework addresses a critical need in autonomous AI workflows:

• Agentic Autonomy: Enables AI tools to generate assets without human intervention
• Comprehensive Asset Types: Supports 2D images, 3D models, audio, and more through a single interface
• Workflow Integration: Seamlessly integrates into AI agent pipelines for end-to-end asset creation
• Backend Flexibility: Works with any asset generation backend, from paid APIs to self-hosted solutions

Features

This server exposes powerful AI asset generation capabilities as MCP tools, allowing AI agents to:

• Universal Asset Generation: Generate 2D images, 3D models, and audio assets through a consistent interface
• Multi-Backend Support: Seamlessly switch between different AI asset generation backends
• Text to 3D: Generate 3D models from text descriptions
• Image to 3D: Convert 2D images into 3D models
• Multi-Image to 3D: Create higher quality models from multiple angles
• Asset Processing: Remeshing, retexturing, rigging, and animation capabilities
• Job Management: Track progress and retrieve results across different backends
• Backend Abstraction: Work with any asset generation API through standardized tools

Supported Backends

• Meshy: Commercial 3D asset generation API
• ComfyUI: Self-hosted AI workflow system
• Stable Diffusion: Image generation capabilities
• And more: Easily extensible to support additional asset generation backends

Prerequisites

• Node.js (v18 or higher)
• Backend-specific credentials: API keys or configuration for your chosen asset generation backends

Installation

1. Clone and Build

git clone <repository-url>
cd asset-mcp
npm install
npm run build

2. Configure Claude Desktop

Add the server to your Claude Desktop configuration file:

Windows (%APPDATA%\Claude\claude_desktop_config.json):

{
  "mcpServers": {
    "asset-mcp": {
      "command": "node",
      "args": ["C:/path/to/asset-mcp/dist/index.js"],
      "env": {
        "MESHY_API_KEY": "your_meshy_api_key_here",
        "COMFYUI_ENDPOINT": "http://localhost:8188"
      }
    }
  }
}

macOS (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "asset-mcp": {
      "command": "node",
      "args": ["/path/to/asset-mcp/dist/index.js"],
      "env": {
        "MESHY_API_KEY": "your_meshy_api_key_here",
        "COMFYUI_ENDPOINT": "http://localhost:8188"
      }
    }
  }
}

Note: Replace /path/to/asset-mcp with the actual absolute path to the project directory.

Development

• Build: npm run build
• Dev (Watch Mode): npm run dev
• Test: npm test
• Test-Harness: npm run test-harness
• Lint: npm run lint
• Format: npm run format

Tools Reference

Generation Tools

• create_text_to_3d

  • prompt: Description of the object.
  • art_style: 'realistic' or 'sculpture' (optional).
  • mode: 'preview' (fast) or 'refine' (high quality).
  • ai_model: 'meshy-4', 'meshy-5', or 'latest'.

• create_image_to_3d

  • image_url: URL of the input image.
  • enable_pbr: Enable PBR maps (optional).
  • should_remesh: Auto-remesh the result (optional).

• create_multi_image_to_3d

  • image_urls: List of 1-4 image URLs.

Processing Tools

• create_remesh

  • model_url or input_task_id: Source model.
  • target_polycount: Desired polygon count.
  • target_formats: Output formats (glb, fbx, obj, etc.).

• create_retexture

  • model_url or input_task_id: Source model.
  • text_style_prompt: Description of new texture style.
  • image_style_url: Reference image for style (optional).

• create_rigging

  • model_url or input_task_id: Source humanoid model.

• create_animation

  • rig_task_id: ID of a completed rigging task.
  • action_id: Animation ID from Meshy library.

Utility Tools

• get_job: Check the status and get results of any task.
• get_balance: Check your current Meshy credit balance.

Extending to New Backends

The architecture is designed to be easily extensible. To add support for a new asset generation backend:

1. Create a new provider in the src/providers/ directory
2. Implement the required interface methods
3. Register the provider in the main configuration
4. The new backend will automatically be available through the same unified tools

License

ISC