MLflow Plugin Manager

PyPI package management integrated directly into MLflow's web interface

Quick Start

Install from PyPI (Recommended)

pip install mlflow-plugin-manager

Install from Source

git clone https://github.com/thijsdezoete/mlflow-plugin-manager.git
cd mlflow-plugin-manager
pip install -e .

Docker

# Using docker compose
docker compose up -d

# Or using make
make docker-build
make docker-run

Core Features

• Direct PyPI Integration - Install any MLflow-compatible package from PyPI through the web interface
• Version Management - Compare installed versions against latest PyPI releases, update or rollback as needed
• Version Pinning - Install specific package versions for reproducibility
• Package Discovery - Browse and search MLflow packages indexed from PyPI
• Installation Logs - View complete pip output for debugging
• REST API - Programmatic access for CI/CD integration

Usage

Run the MLflow server with the Plugin Manager:

mlflow server --app-name plugin_manager

Access the Plugin Manager at:

• MLflow UI: http://localhost:5000
• Plugin Manager: http://localhost:5000/plugin-manager/
• Click "Plugins" in the sidebar for quick access

Configuration

Environment Variables

• PLUGIN_SERVER_URL: URL of the plugin metadata server
  • Default: https://api.mlflowplugins.com (production)
  • Local Development: Set to http://localhost:5001 if running your own server

# Use production server (default)
mlflow server --app-name plugin_manager

# Use local server for development
export PLUGIN_SERVER_URL="http://localhost:5001"
mlflow server --app-name plugin_manager

Make Commands

make help         # Show all available commands

# Development
make install      # Install in development mode
make dev          # Run MLflow with plugin manager
make test         # Run test suite
make lint         # Run linting

# Docker
make docker-build # Build Docker image
make docker-run   # Start container
make docker-stop  # Stop container

# Testing MLflow Versions
make test-mlflow2 # Test with MLflow 2.x
make test-mlflow3 # Test with MLflow 3.x

Architecture

The MLflow Plugin Manager consists of two components:

1. MLflow Plugin (this package): Integrates with MLflow's web interface and handles package installation locally via pip
2. Metadata Server: Provides plugin information (hosted at api.mlflowplugins.com)

Note: Package installation happens locally on your machine. The remote server only provides plugin metadata (available packages, versions, etc.).

API Endpoints

• GET /plugin-manager/ - Web interface
• GET /plugin-manager/installed-plugins - List installed packages
• GET /plugin-manager/available-plugins - Browse PyPI packages
• POST /plugin-manager/install-plugin - Install a package
• POST /plugin-manager/uninstall-plugin - Remove a package
• GET /plugin-manager/check-plugin-updates - Check for updates
• POST /plugin-manager/upgrade-plugin - Update a package

Requirements

• Python 3.10+
• MLflow 2.0+ (fully compatible with MLflow 3.x)
• pip package manager

MLflow Version Compatibility

MLflow Version	Server Type	Status
2.x	Flask/Gunicorn	Supported
3.x	FastAPI/Uvicorn	Supported

The plugin auto-detects the MLflow version and adapts accordingly.

Advanced: Running Your Own Metadata Server

For maintainers who want to run their own plugin metadata server:

cd server
python reindex_plugins.py  # Index packages from PyPI
python app.py              # Run metadata server on port 5001

Note: End users don't need to run these commands - the default configuration uses api.mlflowplugins.com.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

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

Links

• GitHub Repository
• PyPI Package
• Issue Tracker
• MLflow Documentation