MCP Server Replicate

A FastMCP server implementation for the Replicate API, providing resource-based access to AI model inference with a focus on image generation.

Features

• 🖼️ Resource-based image generation and management
• 🔄 Real-time updates through subscriptions
• 📝 Template-driven parameter configuration
• 🔍 Comprehensive model discovery and selection
• 🪝 Webhook integration for external notifications
• 🎨 Quality and style presets for optimal results
• 📊 Progress tracking and status monitoring
• 🔒 Secure API key management

Available Prompts

The server provides several specialized prompts for different tasks:

Text to Image (Primary)

Our most thoroughly tested and robust prompt. Optimized for generating high-quality images from text descriptions with:

• Detailed style control
• Quality presets (draft, balanced, quality, extreme)
• Size and aspect ratio customization
• Progress tracking and real-time updates

Example:

Create a photorealistic mountain landscape at sunset with snow-capped peaks, quality level: quality, style: photorealistic

Other Prompts

• Image to Image: Transform existing images (coming soon)
• Model Selection: Get help choosing the right model for your task
• Parameter Help: Understand and configure model parameters

Prerequisites

• Python 3.11 or higher
• A Replicate API key (get one at https://replicate.com/account)
• UV for dependency management

Installation

Installing via Smithery

To install MCP Server Replicate for Claude Desktop automatically via Smithery:

npx -y @smithery/cli install @gerred/mcp-server-replicate --client claude

Installing Manually

You can install the package directly from PyPI:

# Using UV (recommended)
uv pip install mcp-server-replicate

# Using UVX for isolated environments
uvx install mcp-server-replicate

# Using pip
pip install mcp-server-replicate

Claude Desktop Integration

1. Make sure you have the latest version of Claude Desktop installed
2. Open your Claude Desktop configuration:

# macOS
code ~/Library/Application Support/Claude/claude_desktop_config.json

# Windows
code %APPDATA%Claudeclaude_desktop_config.json

3. Add the server configuration using one of these options:

{
  "globalShortcut": "Shift+Alt+A",
  "mcpServers": {
    "replicate": {
      "command": "uv",
      "args": ["tool", "run", "mcp-server-replicate"],
      "env": {
        "REPLICATE_API_TOKEN": "APITOKEN"
      },
      "cwd": "$PATH_TO_REPO"
    }
  }
}

4. Set your Replicate API key:

# Option 1: Set in your environment
export REPLICATE_API_TOKEN=your_api_key_here

# Option 2: Create a .env file in your home directory
echo "REPLICATE_API_TOKEN=your_api_key_here" > ~/.env

5. Restart Claude Desktop completely

You should now see the 🔨 icon in Claude Desktop, indicating that the MCP server is available.

Usage

Once connected to Claude Desktop, you can:

1. Generate images with natural language:

   Create a photorealistic mountain landscape at sunset with snow-capped peaks
   

2. Browse your generations:

   Show me my recent image generations
   

3. Search through generations:

   Find my landscape generations
   

4. Check generation status:

   What's the status of my last generation?
   

Troubleshooting

Server not showing up in Claude Desktop

1. Check the Claude Desktop logs:

tail -n 20 -f ~/Library/Logs/Claude/mcp*.log

2. Verify your configuration:

• Make sure the path in claude_desktop_config.json is absolute
• Ensure UV is installed and in your PATH
• Check that your Replicate API key is set

3. Try restarting Claude Desktop

For more detailed troubleshooting, see our Debugging Guide.

Documentation

• Implementation Plan
• Contributing Guide
• API Reference
• Resource System
• Template System

Development

1. Clone the repository:

git clone https://github.com/gerred/mcp-server-replicate.git
cd mcp-server-replicate

2. Install development dependencies:

uv pip install --system ".[dev]"

3. Install pre-commit hooks:

pre-commit install

4. Run tests:

pytest

Contributing

We welcome contributions! Please see our Contributing Guide for details.

License

This project is licensed under the MIT License - see the LICENSE file for details.