
Using an engine to choose your moves against a human opponent is cheating. On chess.com and Lichess it is detected and it gets your account closed — the fair play systems look for exactly this pattern, and “I only checked a few moves” is not a defence.
Fine, and what this is built for:
Not fine, and we will not help with it: a live game against another person, anywhere — online, over the board, or correspondence. If you ask Claude for a move in a game against a human, it is instructed to ask who the opponent is and to decline while the answer is a person.
Use the Claude desktop app — that is where the live coach runs. A connector is saved to your Claude account, so it also shows up on claude.ai and on your phone, but the desktop app is the one you want.
https://mcp.betterchess.coThat is the whole setup. Nothing installs on your machine, and subscribing later lifts the cap on the connector you already added — you never reconnect it.
This is the thing in the video on the homepage, and you can run it exactly the same way. Claude watches the board in its own browser, warns you about the captures and checks that lose before you play them, draws the recommendation on the board, and explains every move. Nothing installs.
chess.com/play/computer, pick a bot, press Play. Guest is fine — you do not need a chess.com account.Coach my live chess game against a bot, in the Claude browser.
Claude loads the coaching rig into the page, confirms it is a bot game, and starts. From then on: it waits for the board, analyses your position, draws the move, and tells you why. Press H or the Ask coach button any time to ask about the position. Ask “what about Ng5?” and it tells you what that walks into.
The coach prompt above works on its own. The skill makes it reliable: it tells Claude exactly how to run the loop — arm the rig once, read the board from the page, never stop to ask “say go”, never fall back to a local engine, and explain theplan behind each move (a trade that removes their attacker, a sacrifice and what it buys, a real material win) rather than just naming it.
/plugin marketplace add yuvalsuede/betterchess-connector and then /plugin install betterchess@betterchess. The plugin includes the connector and the skill.Bot games only. If the page is a game against a person, Claude refuses before the first analysis — it reads the page, not what you tell it. Desktop app only for now: the loop needs Claude's own browser, which the web and mobile apps do not have.
There are no commands. Ask in plain language and Claude picks the right tool. These work as-is:
Here is my position: rnbqkbnr/pppp1ppp/8/4p3/4P3/5N2/PPPP1PPP/RNBQKB1R b KQkq - 1 2. What is undefended, what attacks it, and what are the best moves?
Runs chess_analyse. You get the top moves with real evaluations, whose turn it is, whether that side is in check, undefended pieces and what attacks them, and the material balance.
In that position, exactly which squares does the knight on f3 attack?
Runs chess_attacks. This is the one that stops confident nonsense about knights, pins and discovered attacks — the squares are computed, not recalled.
Here is my PGN from a game I just finished. Where did I actually lose it? [paste PGN]
Runs chess_review (Pro). It replays the game and reports only the moves where the evaluation genuinely swung — the two moves that decided it, not forty small inaccuracies.
I am studying this position and I want to play Ng5. Is that actually good, or is there a refutation? [paste FEN]
Analyses the position after your move, so you get the concrete refutation rather than an opinion. Use this while studying — not mid-game against a person.
| Tool | Returns | Plan |
|---|---|---|
chess_analyse | Top moves with evaluations, side to move, check, undefended pieces and their attackers, material balance. | Free (daily budget) |
chess_attacks | The exact squares a given piece attacks, and its type and colour. | Free (daily budget) |
chess_review | Every move in a finished game where the evaluation swung past a threshold, with the move played, the better move, and the position after. | Pro |
No. Paste the connector URL into Claude's connector settings and sign in with your BetterChess account. There is no terminal step, no config file and no local Stockfish. It works on the free Claude plan and on mobile.
There is a coach prompt in the connector, and a chess-live-coach skill you can install (download it from this page, or install the BetterChess plugin in Cowork / Claude Code). With the skill on, saying "coach me" with a bot game open is enough.
No. Using an engine to pick moves against a human opponent is cheating and will get your account banned on chess.com and Lichess. Against a bot, in a finished game, or on a position you are studying, it is fine — see the fair play section above.
A free account gets a daily analysis budget, which is enough to work through a game or study positions. Pro removes the cap for $5.99/month, cancel anytime.
The board facts do not. Attackers, defenders, evaluations and legal moves are computed by the engine and handed to Claude as fact. Claude still writes the explanation around them, so if it ever says something that contradicts the numbers it returned, tell it — it will re-check.
FEN, PGN, or a plain list of moves from the start (e.g. e4 e5 Nf3 Nc6). Paste whichever you have.
Questions or something not working: info@betterchess.co. BetterChess is independent and not affiliated with Chess.com, Lichess or Anthropic; those names are their owners' trademarks. BetterChess is a practice tool — we make no guarantee of any rating result.
PRINTED CHESS PUZZLE CARDS
54 poker-size cards in a tuck box. Puzzle on the front, solution on the back.