I wanted to learn chess but never started because the feedback cycle took too long. Play on chess.com, go study moves somewhere else, come back, and you probably wont even see those moves in your next game. The loop never closes.

So I built Chess Rocket to hack that. It connects Claude Code to Stockfish through MCP, and Claude becomes your coach in real time. Every move gets evaluated while your playing. Blunders get full explanations, good moves get a nod. The depth of teaching scales with how bad the move was.

The key pieces: adaptive difficulty that adjusts to your level, spaced repetition (SM-2) that creates flashcards from your mistakes automatically, 3500+ openings from Lichess with live identification, and 280+ curated puzzles. The repetition and behavioral analysis is all handled by the system. You just play chess and talk to Claude, thats it.

I've been using it for 12 sessions, went from ~400 to ~700 Elo. The SRS cards do alot of the heavy lifting honestly.

Open source: https://github.com/suvojit-0x55aa/chess_rocket

Happy to answer anything about the teaching methodology or architecture.