mcpField
M
Meshy AI MCP Server

Meshy AI MCP Server

Local
github.com/pasie15Media & Design43Updated 10w ago

About Meshy AI MCP Server

Meshy AI MCP Server connects MCP clients to Meshy AI's generative media API. It covers 3D creation and editing alongside image generation, with task tracking and progress streaming for long-running operations.

Key features

  1. 01Generate 3D models from text or images
  2. 02Create models from multiple reference images
  3. 03Retexture and remesh models
  4. 04Generate and transform images
  5. 05Rig and animate 3D characters
  6. 06Track tasks and stream progress

Use cases

  1. 01Create 3D assets from text descriptions or references
  2. 02Revise textures or optimize model geometry
  3. 03Rig a character and apply an animation

MCP support

Exposes Meshy AI API operations as tools for MCP clients, including model and image generation, editing, rigging, animation, task management, and balance lookup. Calls require a Meshy AI API key.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

No install method listed yet

We haven’t collected a package or remote endpoint for this server. Follow the instructions in the project README.

Repository

Tools

No tool list yet

Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.

Details

From the project README

Meshy AI MCP Server

This is a Model Context Protocol (MCP) server that wraps the Meshy AI API. It enables MCP clients (like Claude Desktop, Cursor, Cline) to interact with Meshy's generative 3D tools directly.

Features

  • Text-to-3D: Generate 3D models from text prompts.
  • Image-to-3D: Create 3D models from reference images.
  • Multi-Image-to-3D: Create 3D models from multiple reference images.
  • Retexture: Apply new textures to existing 3D models. The text-to-texture tools are compatibility aliases for the same retexture API.
  • Text-to-Image: Generate images from text prompts.
  • Image-to-Image: Generate new images from input images.
  • Model Optimization: Remesh and optimize geometry.
  • Rigging: Auto-rig 3D characters for animation.
  • Animation: Apply animations to rigged characters.
  • Streaming: Real-time progress updates for long-running tasks.
  • Task Deletion: Delete tasks across all API categories.

Installation

Option 1: Run directly with npx (Recommended)

You can run the server directly using npx without installing it globally.

{
  "mcpServers": {
    "meshy-ai": {
      "command": "npx",
      "args": [
        "-y",
        "meshy-ai-mcp-server"
      ],
      "env": {
        "MESHY_API_KEY": "your_meshy_api_key_here"
      }
    }
  }
}
Option 2: Clone and Build Locally

If you want to modify the code or run it from a local source:

  1. Clone the repository:

    git clone <repository-url>
    cd meshy-ai-mcp-server
    
  2. Install dependencies:

    npm install
    
  3. Build the project:

    npm run build
    
  4. Configure your MCP Client:

    Add the following to your MCP client configuration (e.g., claude_desktop_config.json or VS Code settings):

    {
      "mcpServers": {
        "meshy-ai": {
          "command": "node",
          "args": [
            "/absolute/path/to/meshy-ai-mcp-server/dist/index.js"
          ],
          "env": {
            "MESHY_API_KEY": "your_meshy_api_key_here"
          }
        }
      }
    }
    

Configuration

You need a Meshy AI API key to use this server.

  1. Get your API key from the Meshy Dashboard.
  2. Set the MESHY_API_KEY environment variable in your MCP client configuration (as shown above).
Optional Environment Variables
  • MESHY_API_BASE: Override the API base URL (default: https://api.meshy.ai/openapi).
  • MESHY_STREAM_TIMEOUT_MS: Timeout for streaming responses in milliseconds (default: 300000 aka 5 minutes).
Troubleshooting

If your MCP client reports that the server closed during initialize, check that the client configuration passes MESHY_API_KEY into the server process. The server can start without the key so clients can inspect available tools, but Meshy API tool calls will fail until the key is configured.

Development

To run the server in development mode with auto-reloading:

# Create a .env file
echo "MESHY_API_KEY=your_key_here" > .env

# Run in dev mode
npm run dev

Available Tools

  • Text to 3D: create_text_to_3d_task, retrieve_text_to_3d_task, list_text_to_3d_tasks, stream_text_to_3d_task, delete_text_to_3d_task
  • Image to 3D: create_image_to_3d_task, retrieve_image_to_3d_task, list_image_to_3d_tasks, stream_image_to_3d_task, delete_image_to_3d_task
  • Multi-Image to 3D: create_multi_image_to_3d_task, retrieve_multi_image_to_3d_task, list_multi_image_to_3d_tasks, stream_multi_image_to_3d_task, delete_multi_image_to_3d_task
  • Retexture: create_retexture_task, retrieve_retexture_task, list_retexture_tasks, stream_retexture_task, delete_retexture_task
  • Texturing (retexture aliases): create_text_to_texture_task, retrieve_text_to_texture_task, list_text_to_texture_tasks, stream_text_to_texture_task, delete_text_to_texture_task — same retexture API, historical tool names kept for compatibility.
  • Text to Image: create_text_to_image_task, retrieve_text_to_image_task, list_text_to_image_tasks, stream_text_to_image_task, delete_text_to_image_task
  • Image to Image: create_image_to_image_task, retrieve_image_to_image_task, list_image_to_image_tasks, stream_image_to_image_task, delete_image_to_image_task
  • Remeshing: create_remesh_task, retrieve_remesh_task, list_remesh_tasks, stream_remesh_task, delete_remesh_task
  • Rigging: create_rigging_task, retrieve_rigging_task, list_rigging_tasks, stream_rigging_task, delete_rigging_task
  • Animation: create_animation_task, retrieve_animation_task, list_animation_tasks, stream_animation_task, delete_animation_task
  • Utility: get_balance

License

MIT

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.