If you've ever played a competitive TCG—Magic: The Gathering, Pokémon, Lorcana—you know the feeling. You spend three hours tweaking a list. You add one more land here, swap a creature for a spell there. Then the game starts, and within two turns, you realize your deck is fundamentally broken because you missed your third land drop or drew nothing but high-cost threats.
The traditional way to fix this involves spreadsheets, manual hypergeometric calculations, and praying you didn't mess up the formula in Excel. Or worse, asking an LLM 'Is my deck good?'
Asking an LLM if your deck is good is dangerous territory. Most models will hallucinate a sense of confidence based on keywords. They might tell you 24 lands in a 60-card deck feels 'balanced' because they've seen similar text during training. But 'feels balanced' isn't a mathematical proof. It doesn't account for the specific variance of hitting Turn 3 consistently versus Turn 4.
I wanted to bridge this gap between probabilistic certainty and agentic reasoning. This led me to develop the TCG Mana Curve Analyzer, an MCP server designed specifically to move the heavy lifting of combinatorial mathematics away from the model's weights and into a deterministic engine.
