splatter -app

This MCP server hosts the Splatter app, which enables generating and viewing 3D models using 3D Gaussian Splatting. The server exposes tooling to upload and render 3D content and to convert images into splats via a Modal-backed pipeline. End users can upload .ply models to render in an interactive widget, or supply an image to trigger an image-to-splat workflow through the provided tools. The app integrates with ChatGPT-generated prompts to demonstrate a live image-to-3D workflow, making it suitable for rapid prototyping in advertising, media, creative design, architecture, and real estate contexts. The main entry points you’ll interact with are open-ply-upload (to bring in new meshes), view-ply-splat (to render an existing .ply), and generate-splat-from-image (to convert images to splats).

Prerequisites:

• Node.js and pnpm installed on your development machine
• Git available to clone the repository

1. Clone the repository

git clone https://github.com/your-org/splatter-mcp-app.git
cd splatter-mcp-app

2. Install dependencies

pnpm install

3. Run the development server

pnpm run dev

Optional for tunnel testing (if you want to expose locally):

pnpm run dev -- --tunnel

Notes:

• Ensure SHARP_MODAL_ENDPOINT_URL is set to a live deployed endpoint before using image-to-splat features.
• The Modal backend is deployed under services/sharp-inference and can be started with the provided uv commands in the README; ensure uv is installed for the Python backend.
• If you’re running locally, configure MCP_URL to point to your public URL or tunnel endpoint to enable external access.
• The app uses pnpm scripts; you can adapt to npm or yarn if needed, but consistency is recommended.
• For quality checks, run the project’s lint/typing checks as shown in the README (ruff, type checking, etc.).