Vybe Virtual Try-On MCP Server

A FastMCP server that wraps the Replicate API for virtual try-on functionality, ready for deployment to Render.

Setup

Using UV (recommended)

1. Install uv if you haven't already:

2. Install dependencies:

3. Set up environment variables:

Edit .env and add your Replicate API token.

Using pip (alternative)

1. Install dependencies:

2. Set up environment variables:

Edit .env and add your Replicate API token.

Related MCP server: MCP Server for Replicate

Local Testing

Run the server locally:

The server will start on the default FastMCP port and expose the virtual_tryon tool.

MCP Client Configuration

Add to your Claude Desktop or other MCP client configuration:

Usage

The server exposes three tools:

test_connection

Tests the connection and shows timeout configuration.

base64_to_url

Converts base64 encoded images to data URIs for use with virtual_tryon:

• base64_image: Base64 encoded image string (with or without data:image prefix)

• image_type: Image type (png, jpg, jpeg, gif, webp) - default: png

Returns a data URI that can be used as model_image or garment_image in virtual_tryon.

virtual_tryon

Performs virtual try-on with these parameters:

• model_image: URL or data URI of the person/model image

• garment_image: URL or data URI of the clothing item to try on

• Various optional parameters for customization

Timeout Configuration

The server is configured with extended timeouts to handle long-running Replicate operations:

• MCP request timeout: 600 seconds (10 minutes)

• Replicate polling interval: 5 seconds

• Replicate timeout: 600 seconds (10 minutes)

If you still experience timeouts, you can adjust these in the server.py file.

Deployment to Render

Quick Deploy with render.yaml

1. Push your code to a GitHub repository

2. Connect the repository to Render

3. Render will automatically detect the render.yaml configuration

4. Add your REPLICATE_API_TOKEN in the Render dashboard under Environment Variables

5. Deploy!

The service includes health check endpoints:

• /health - Health check endpoint for Render monitoring

• / - Root endpoint with service status

Manual Setup

If you prefer manual configuration:

1. Create a new Web Service on Render

2. Connect your GitHub repository

3. Configure the service:

   • Build Command: curl -LsSf https://astral.sh/uv/install.sh | sh && source $HOME/.cargo/env && uv sync --frozen --no-dev

   • Start Command: uv run python server.py

   • Health Check Path: /health

4. Add environment variables:

   • REPLICATE_API_TOKEN: Your Replicate API token (required)

   • PORT: (leave empty, Render will auto-assign)

   • HOST: (leave empty, defaults to 0.0.0.0)

Using the Remote MCP Server

Once deployed, you can access the server at:

• Health Check: https://your-service-name.onrender.com/health

• Root: https://your-service-name.onrender.com/

• MCP Protocol: Use the base URL for MCP client connections

For MCP clients that support HTTP transport:

Replace your-service-name with your actual Render service URL.

Testing the Deployment

Use the included test script to verify your deployment is working:

This will test:

• Health check endpoint (/health)

• Root endpoint (/)

• Basic MCP server connectivity

Example output:

Docker Deployment (Alternative)

A Dockerfile is also included if you prefer containerized deployment. Render will automatically detect and use it if present.