Leonardo AI

A Model Context Protocol (MCP) server for Leonardo AI, supporting both HTTP and stdio modes.

Features

• Create image generation jobs with Leonardo AI
• See available models
• Check the status of image generation jobs
• Get all the user's image generation jobs
• Supports both HTTP and stdio transports

Installation

JSON Config

Support for Claude Desktop, Cursor and other MCP clients that use JSON config files.

[!IMPORTANT]
You will need to generate a Leonardo API key and set it in the environment variable LEONARDO_API_KEY before running the server.

{
  "mcpServers": {
    "leonardo-mcp-server": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/ish-joshi/leonardo-mcp-server",
        "leonardo-mcp-server",
        "stdio"
      ],
      "env": {
        "LEONARDO_API_KEY": "YOUR_LEONARDO_API_KEY"
      }
    }
  }
}

Demo

IMAGE ALT TEXT HERE

Running Modes

This server supports two modes:

• HTTP mode (default):
  • Suitable for remote clients (e.g., ChatGPT Playground, browser-based tools).
  • The server runs an HTTP endpoint. You must expose it to the internet if your client is remote.
• Stdio mode:
  • Suitable for local clients that communicate over standard input/output (stdio).
  • No network port is opened.

HTTP Mode (for remote clients)

Start the server in HTTP mode (default):

uvx --from git+https://github.com/ish-joshi/leonardo-mcp-server leonardo-mcp-server

If your client is remote (e.g., ChatGPT Playground), you must expose your local server to the internet. You can use ngrok or a similar tunneling tool:

ngrok http 8080

Copy the public URL from ngrok and use it as the endpoint in your client.

Stdio Mode (for local clients)

Start the server in stdio mode:

uvx --from git+https://github.com/ish-joshi/leonardo-mcp-server leonardo-mcp-server stdio

Environment Variables

• LEONARDO_API_KEY (required): Your Leonardo AI API key.

Development

• Edit main.py to add or modify MCP tools.
• Run using python main.py and test with a compatible MCP client. I prefer to use 5ire MCP client for testing.
• See python-sdk documentation for more info.