VGGT-MPS

Provides Apple Silicon-accelerated multi-view 3D reconstruction with MCP integration for Claude Desktop.

python

79

GitHub Stars

python

Language

6 months ago

First Indexed

2 months ago

Catalog Refreshed

Documentation & install

Readme and setup notes from the catalogue, plus a client-ready config you can copy for your MCP host.

Installation

Add the following to your MCP client configuration file.

Configuration

View docs

{
  "mcpServers": {
    "jmanhype-vggt-mps": {
      "command": "uv",
      "args": [
        "run",
        "--python",
        "/path/to/vggt-mps/vggt-env/bin/python",
        "--with",
        "fastmcp",
        "fastmcp",
        "run",
        "/path/to/vggt-mps/src/vggt_mps_mcp.py"
      ],
      "env": {
        "VGGt_ENV": ".env file contents"
      }
    }
  }
}

VGGT-MPS MCP Server enables fast, GPU-accelerated 3D reconstruction on Apple Silicon and provides a ready-to-use integration path with MCP clients. It exposes a compact, scriptable interface to run multi-view 3D reconstruction powered by VGGT, optimized for MPS on M1/M2/M3 Macs, and makes it easy to connect to MCP-enabled environments such as Claude Desktop.

How to use

You connect to this MCP server from your MCP client to perform 3D reconstruction tasks on Apple Silicon. Use the provided integration entry to run the agent locally and wire it into your MCP workflow. The agent exposes the VGGT processing pipeline with MPS acceleration, multi-view reconstruction, and export options.

How to install

Prerequisites You need an Apple Silicon Mac (M1/M2/M3) with Python 3.10 or newer and a reasonable amount of RAM (8GB+). Ensure you have an MCP-capable client to talk to the server.

Step 1: Prepare runtime environment Ensure Python 3.10+ is installed on your Mac and that you can run Python commands from your shell.

Step 2: Obtain MCP integration configuration Copy the MCP configuration snippet below and place it into your Claude Desktop MCP configuration file or the equivalent MCP client configuration area.

Step 3: Run the integration command Use the exact runtime command and arguments shown in the configuration snippet to start the MCP agent when you want to connect Claude Desktop to VGGT-MPS.

Step 4: Verify and test After configuring, test the integration by launching Claude Desktop and confirming that the VGGT-MPS agent starts successfully and is reachable through the MCP interface.

MCP integration configuration

{
  "mcpServers": {
    "vggt-agent": {
      "command": "uv",
      "args": [
        "run",
        "--python",
        "/path/to/vggt-mps/vggt-env/bin/python",
        "--with",
        "fastmcp",
        "fastmcp",
        "run",
        "/path/to/vggt-mps/src/vggt_mps_mcp.py"
      ]
    }
  }
}

Additional sections

This MCP server is designed to work with the VGGT-MPS toolchain. It includes a unified entry point for the 3D reconstruction workflow, MPS acceleration, and multiple export formats (PLY, OBJ, GLB). Other useful capabilities include depth maps, camera poses, 3D point clouds, and confidence maps generation from multi-view imagery.

Tools and endpoints

Available VGGT-related MCP tools exposed by this server include quick start inference, video frame extraction, image processing, and 3D scene creation and visualization. Each tool provides a practical, end-to-end capability for building and inspecting 3D reconstructions from images.

Available tools

vggt_quick_start_inference

Quick 3D reconstruction from images using the VGGT-MPS pipeline.

vggt_extract_video_frames

Extract frames from a video to feed into the VGGT pipeline.

vggt_process_images

Full VGGT processing pipeline for multi-view images.

vggt_create_3d_scene

Generate GLB/PLY/OBJ representations of 3D scenes.

vggt_reconstruct_3d_scene

Multi-view reconstruction to build depth maps, poses, and point clouds.

vggt_visualize_reconstruction

Create visualizations of the reconstructed 3D scene.