google-deepmind/mctx

Health v2 · maintenance only

Adoption not part of the score

2654 stars · 215 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Mctx is a JAX-native Python library implementing Monte Carlo tree search algorithms such as AlphaZero, MuZero, and Gumbel MuZero. It supports JIT compilation and batched parallel search on accelerators, making it suitable for research on search-based reinforcement learning agents.

Use cases

• implement mcts in jax
• build a muzero agent
• run alphazero-style search with neural networks
• batched parallel tree search on gpu
• research planning algorithms in python
• combine learned models with tree search

When to choose

• you need MCTS integrated with JAX-based neural network training
• you want hackable, pure-Python search implementations instead of C++
• you need batched, JIT-compiled search for accelerator efficiency

When to avoid

• you need a ready-made game engine or complete RL agent rather than search primitives
• your project does not use JAX
• you need CPU-only classical MCTS for board games without learned models

Facets

library · maturity active

reinforcement-learning machine-learning simulation reinforcement-learning machine-learning deep-learning python cross-platform mcts jax muzero alphazero planning search-algorithms jit-compilation algorithms gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="google-deepmind/mctx")