Mlflowmcpserver

🚀 MLflow MCP Server: Machine Learning Workflow with Natural Language Interface

This project offers a natural language interface to access MLflow via the Model Context Protocol (MCP). It allows you to query your MLflow tracking server using simple English, making it easier to manage and explore your machine learning experiments and models.

🚀 Quick Start

MLflow MCP Proxy consists of two main components:

1. MLflow MCP Server (mlflow_server.py): Connects to your MLflow tracking server and exposes MLflow capabilities through the Model Context Protocol (MCP).
2. MLflow MCP Client (mlflow_client.py): Provides a natural language interface and uses a conversational AI assistant to interact with the MLflow MCP Server.

✨ Features

• Natural Language Queries: Ask questions about your MLflow tracking server in simple English.
• Model Registry Exploration: Obtain information about your registered models.
• Experiment Tracking: List and explore your experiments and runs.
• System Information: Get the status and metadata of the MLflow environment.

📦 Installation

Prerequisites

• Python 3.8+
• A running MLflow tracking server (default: http://localhost:8080)
• OpenAI API key for LLM

Steps

1. Clone this repository:

   git clone https://github.com/iRahulPandey/mlflowMCPServer.git
   cd mlflowMCPServer
   

2. Create a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate
   

3. Install the required packages:

   pip install mcp[cli] langchain-mcp-adapters langchain-openai langgraph mlflow
   

4. Set your OpenAI API key:

   export OPENAI_API_KEY=your_key_here
   

5. (Optional) Configure the MLflow tracking server URI:

   export MLFLOW_TRACKING_URI=http://localhost:8080
   

💻 Usage Examples

Starting the MCP Server

First, start the MLflow MCP Server:

python mlflow_server.py

This server connects to your MLflow tracking server and exposes MLflow capabilities through MCP.

Making Queries

Once the server is running, you can use the client to make natural language queries:

python mlflow_client.py "What models are registered in my MLflow?"

Example queries:

• "Show all models registered in MLflow."
• "List all my experiments."
• "Get details of the model named 'iris-classifier'."
• "What's the status of my MLflow server?"

📚 Documentation

Configuration

You can customize the behavior using environment variables:

MLflow MCP Server (mlflow_server.py)

This server connects to your MLflow tracking server and exposes the following capabilities through MCP:

• /api/v1/predict: Handles prediction requests.
• /api/v1/models: Lists all registered models. To use these APIs, send requests to http://localhost:5000 as the server runs on port 5000.

Limitations

The current version only supports the following features:

• Listing registered models
• Performing predictions

The following operations are not supported:

• Model upload
• Model versioning
• Experiment tracking

Example Usage

To use the MLflow MCP Client, follow these steps:

1. Start the server:

   python mlflow_server.py
   

2. Use the client to make a query:

   python mlflow_client.py "Show all models registered in MLflow"
   

📄 Issues

If you encounter any issues, please submit them to the GitHub repository.