Best Gaming harnesses for AI agents
The most-adopted Gaming harnesses an AI agent can use, ranked by GitHub stars, with what each is best for. Loadbay is an MCP server, so an agent can pull this list live:
claude mcp add --transport http loadbay https://loadbay.xyz/api/mcp
-
1. generative_agents
21,577★ · Python
Most adopted — the default starting point. Best for OpenAI. Stanford 'Smallville' simulation of interactive LLM-driven agents that remember, reflect, and plan in a sandbox town. -
2. ml-agents
19,494★ · C#
Best for Unity, PyTorch. Unity toolkit that turns games and simulations into environments for training agents via RL and imitation learning. -
3. Stable-Baselines3
13,448★ · Python
Best for Gymnasium, PyTorch, ALE. The standard library of reliable PyTorch RL algorithm implementations (PPO, SAC, DQN) used to train game-playing agents. -
4. Gymnasium
12,057★ · Python
Best for Atari, MuJoCo. Standard API and reference environments for single-agent reinforcement learning, the maintained successor to OpenAI Gym. -
5. unity-mcp
10,763★ · C#
Best for Unity. MCP bridge between AI assistants and the Unity Editor for managing assets, controlling scenes, and editing scripts. -
6. pysc2
8,295★ · Python
Best for StarCraft II. DeepMind's StarCraft II learning environment exposing the game to RL agents through a Python observation and action API. -
7. DeepMind Lab
7,365★ · C
Best for Python, Bazel. Customisable 3D first-person platform for agent research, providing navigation, memory, and puzzle tasks for RL agents. -
8. Voyager
6,989★ · JavaScript
Best for Minecraft, Mineflayer. LLM-powered open-ended embodied agent that autonomously explores, learns skills, and plays Minecraft via the Mineflayer bot API. -
9. open_spiel
5,275★ · C++
Best for Chess, Poker, Go. Collection of environments and algorithms for reinforcement learning and search in over 70 board, card, and strategy games. -
10. Google Research Football
3,620★ · Python
Best for OpenAI Gym, TensorFlow, PettingZoo. Physics-based 3D football environment for training RL agents in single- and multi-agent modes with a scenario academy.