solo-3d

3D MCP Tools

Install

pip install pydantic litserve fastmcp open3d

This repository contains three Model Context Protocol (MCP) tools for 3D data processing and analysis:

1. Point Cloud Generation MCP (mcp/mcp_point_cloud_generation.py)

Purpose: Generates 3D point clouds from image sequences using the CUT3R (ARCroco3DStereo) model.

Input:

• seq_path: Path to the image sequence folder
• model_path: Path to the CUT3R model (default: "src/cut3r_512_dpt_4_64.pth")
• size: Image size for processing (default: 512)
• vis_threshold: Visibility threshold for point filtering (default: 1.5)
• skip_frames: Whether to skip frames during processing (default: True)
• output_dir: Output directory path (optional)

Output:

• A PLY file containing the generated 3D point cloud (fused.ply)
• Camera intrinsics and poses for each frame
• Depth maps and confidence for each frame
• Processed image frames
• Returns the path to the generated point cloud file and output folder

Use Case: Convert 2D image sequences (like video frames) into 3D point cloud representations for further processing.

2. Mesh Generation MCP (mcp/mcp_mesh_generation.py)

Purpose: Converts PLY point cloud files into 3D mesh surfaces using various reconstruction algorithms.

Input:

• ply_path: Path to the input PLY point cloud file
• method: Mesh reconstruction method (poisson, alpha, or ball_pivoting)
• output_path: Output path for the generated mesh (optional)
• poisson_depth: Depth parameter for Poisson reconstruction (default: 8)
• alpha: Alpha parameter for alpha shape reconstruction (default: 0.03)
• ball_radii: Ball radii for ball pivoting reconstruction (default: [0.005, 0.01, 0.02])

Output:

• A PLY file containing the generated 3D mesh surface
• Returns the path to the generated mesh file

Methods Available:

• Poisson: Surface reconstruction using Poisson surface reconstruction
• Alpha: Surface reconstruction using alpha shapes
• Ball Pivoting: Surface reconstruction using ball pivoting algorithm

Use Case: Convert sparse point clouds into continuous mesh surfaces for visualization, analysis, or 3D printing.

3. Floor Data MCP (mcp/mcp_floor_data.py)

Purpose: Generates 2D floor maps from PLY point cloud files for navigation and planning.

Input:

• ply_path: Path to the PLY point cloud file
• output_path: Output path for the generated floor map (optional)

Output:

• A 2D floor map representation of the 3D environment
• Returns the generated floor data

Use Case: Create navigable 2D representations of 3D environments for robotics, autonomous navigation, or spatial analysis.

Installation and Usage

Each MCP tool can be run independently as a server on port 8000. The tools automatically clone and set up the required dependencies from their respective repositories:

• Point Cloud Generation: Uses forked CUT3R repository
• Mesh Generation: Uses solo-3d repository
• Floor Data: Uses solo-3d repository

All tools require CUDA support for optimal performance, especially the point cloud generation tool which uses deep learning models.