Mlflow Prompt Registry

🚀 MLflow Prompt Registry MCP Server

The Model Context Protocol (MCP) server is implemented for the MLflow Prompt Registry, enabling users to access prompt templates managed in MLflow.

This server implements the MCP Prompts specification for discovering and using prompt templates hosted in the MLflow Prompt Registry. The primary use case is to load prompt templates from MLflow in Claude Desktop, allowing users to conveniently instruct Claude to perform repetitive tasks or common workflows.

![./assets/mcp-mlflow-prompt-demo.png]

🚀 Quick Start

✨ Features

• list-prompts:
  • Lists available prompts.
  • Inputs:
    • cursor (optional string): Pagination cursor.
    • filter (optional string): Prompt filter.
  • Returns: A list of prompt objects.
• get-prompt:
  • Retrieves and compiles a specific prompt.
  • Inputs:
    • name (string): The name of the prompt to retrieve.
    • arguments (optional object): A JSON object containing prompt variables.
  • Returns: A compiled prompt object.

📦 Installation

Step 1: Install MLflow and Start the Prompt Registry

If not already installed, install and start an MLflow server to host the prompt registry:

pip install mlflow>=2.21.1
mlflow server --port 5000

Step 2: Create a Prompt Template in MLflow

If you haven't already, follow this guide to create a prompt template in MLflow.

Step 3: Build the MCP Server

npm install
npm run build

Step 4: Add the Server to Claude Desktop

Configure Claude for Desktop by editing claude_desktop_config.json:

{
  "mcpServers": {
    "mlflow": {
      "command": "node",
      "args": ["<absolute-path-to-this-repository>/dist/index.js"],
      "env": {
        "MLFLOW_TRACKING_URI": "http://localhost:5000"
      }
    }
  }
}

Make sure to replace MLFLOW_TRACKING_URI with your actual MLflow server address.