A developer has successfully trained large language models to play Magic: The Gathering against each other using MCP tools connected to the open-source XMage codebase. The system works despite being somewhat buggy, demonstrating that AI can handle complex card game mechanics through appropriate tooling. The implementation shows promise for future improvements.

Currently, the ratings for expensive frontier models appear artificially low because development has focused on cheaper models while debugging the system. This means high-end models haven't had enough games in the system to demonstrate their full potential. The developer notes significant room for improvement through better tooling, suggesting the current implementation is just the beginning of what's possible.

The project represents an interesting intersection of AI and gaming, where language models must understand and execute complex rule sets. While still in early stages, this approach could pave the way for AI opponents that understand nuanced game mechanics rather than just following scripted patterns. The use of open-source tools like XMage makes this accessible for further development.