playidea-lab/pcq

What it does

playidea-lab/pcq is an agentoperable ML experiment contract framework that standardizes and manages machine learning experiments through YAML and JSON contract specifications. It exposes 14 MCP tools including resolve, inspect, run, validate, describe, compare, and lineage functions for comprehensive experiment orchestration across PyTorch, Hugging Face Trainer, Lightning, scikit-learn, and XGBoost frameworks.

Key features

• Multi-framework support (PyTorch, HF Trainer, Lightning, sklearn, XGBoost)
• 14 built-in MCP tools for experiment lifecycle management
• Contract-based experiment definition (cq.yaml + JSON)
• Experiment validation and comparison capabilities
• Lineage tracking for reproducibility
• Agent-operable design for automated workflows
• Cross-platform support (Python, macOS, Windows, Linux)

How to set up

Install via GitHub repository with Python environment setup. Clone the repository, install dependencies, configure your experiment contracts in cq.yaml format, and integrate with MCP server for agent-based access. Supports direct framework integration for PyTorch, Hugging Face, Lightning, and scikit-learn projects.