About

cEDH AI is a systems and research project — a fully deterministic Magic: The Gathering rules engine built for Competitive Commander (cEDH). It is not a consumer product. It is a platform.

The engine models the Comprehensive Rules explicitly: turn structure, priority, the stack, state-based actions, continuous effects across all seven dependency layers, replacement effects, triggered and activated abilities, and Commander-specific rules. Cards are not implemented as code — they are compiled from Oracle text into structured JSON that the engine evaluates deterministically.

The long-term goal is a stable game API that AI agents, simulators, and research tools can consume headlessly — no UI required. A React frontend exists for inspection and private playtesting, not as the primary interface.

Current State

  • 4,500+ passing tests, 0 failures
  • 423 hand-curated cEDH staples + 30,000+ auto-compiled cards
  • Five AI agents: Random → Heuristic → Evaluation → MCTS → Combo
  • Full lobby system with multi-player support and WebSocket real-time updates
  • 92% Comprehensive Rules compliance, active development toward 100%

This is private alpha. Rules correctness is the top priority.
Play a game →    Read the research →

Play

Create Game Lobby

Set up a multiplayer session. Share the invite link — each player joins and submits their own decklist before the game starts.

Quick Game

Play immediately against three AI opponents using the default Kenrith vs Kalamax vs Tymna/Thrasios vs Rograkh/Silas 4-player configuration. No setup required.

Custom Game

configure decklists & players

Up to 4 players with custom decklists. Players 1 and 2 are pre-loaded with the Kenrith and Kalamax defaults — edit any textarea before starting.

Research

Published technical writing on game complexity, AI, and formal rules modeling for card games.

Chess vs. Go vs. Magic: Comparative Complexity

A cross-language analysis of computational complexity across three canonical strategy games, with focus on Magic: The Gathering's unbounded state space and ruling interaction depth.