I taught LLMs to play Magic: The Gathering against each other
Details
- External ID
- 47049227
- Source
- HN
- Company
- —
- Product
- I taught LLMs to play Magic: The Gathering against each other
- Website domain
- mage-bench.com
- Launched
- Feb. 17, 2026
- Cohort
- —
- Upvotes
- 117
- Upvotes percentile
- 0.9056603773584906
- Tags
- —
- Fetched at
- Sept. 7, 2026, 9:25 p.m.
- Updated at
- Sept. 7, 2026, 9:25 p.m.
Description
I've been teaching LLMs to play Magic: The Gathering recently, via MCP tools hooked up to the open-source XMage codebase. It's still pretty buggy and I think there's significant room for existing models to get better at it via tooling improvements, but it pretty much works today. The ratings for expensive frontier models are artificially low right now because I've been focusing on cheaper models until I work out the bugs, so they don't have a lot of games in the system.
Enrichment
- Theme
- Vertical
- Media & entertainment
- Function
- Agent / copilot
- Audience
- B2C
- AI stance
- AI-native
- Project type
- Hobby / open-source project
- Normalized one-liner
- llms playing magic the gathering
- Manually corrected
- False
Could you build this?
Yes The system interfaces an existing open-source game engine (XMage) via the Model Context Protocol (MCP) or JSON-RPC to expose game state and actions to LLM API clients.
Discussion
20 comments analyzed.
Competitors mentioned: 17 Lands (for game data collection), AlphaGo-style implementations (for game AI), Ryan Saxe's MTG project
Concerns raised: LLMs struggle with basic arithmetic and hand evaluation, Models hallucinate card effects without proper context, Blunder detection metrics are unreliable and need refinement, Difficulty measuring performance meaningfully, Models lack understanding of format-specific rules and meta context
Feature requests: Add counterfactual simulation via parallel XMage runs with MCTS for deeper analysis, Include price-to-performance comparison metrics, Extend to sideboarding, deck selection, deckbuilding, and drafting, Provide oracle text for all visible cards in mulligan decisions, Match up game IDs across 17 Lands for full hand information datasets
Competitors
Other products that read as similar to this one — 156 launches clear the similarity bar, closest 8 shown.
Attention rank: #19 of 157 (itself plus its competitors, highest first — normalized so YC and Product Hunt are compared fairly).
Launched 109 days after the earliest competitor.
- Cardcast.gg · hn · 2026-04-24 · 5 upvotes · similarity 0.58
- Watch LLMs play 21,000 hands of Poker · hn · 2026-01-08 · 36 upvotes · similarity 0.51
- Play poker with LLMs, or watch them play against each other · hn · 2026-01-10 · 163 upvotes · similarity 0.51
- Shoot-em-up Deck Building game · hn · 2026-01-06 · 5 upvotes · similarity 0.49
- I wrapped the Zorks with an LLM · hn · 2026-01-27 · 111 upvotes · similarity 0.47
- 1v1 coding game that LLMs struggle with · hn · 2026-03-06 · 29 upvotes · similarity 0.46
- A real-time strategy game that AI agents can play · hn · 2026-02-25 · 220 upvotes · similarity 0.45
- Antimatter · hn · 2026-03-24 · 12 upvotes · similarity 0.44
Other launches for this product
- No other launches for this product.
Same idea, different domain
Nobody's really built a agent / copilot tool for Agriculture yet.