# Claude chess skill: what exists and how to use one > Is there a Claude chess skill? What Agent Skills and MCP connectors are, the chess skills on GitHub, and how to use a chess connector in Claude Code. - URL: https://thechesscoach.netlify.app/guides/claude-code-chess-skill/ - Updated: 2026-10-26 - Author: Ignacio Soler There is no official Claude chess skill. As of October 26, 2026, Anthropic's pre-built Agent Skills cover PowerPoint, Excel, Word and PDF, and nothing about chess. What you can find are community chess skills on GitHub, mostly for Claude Code, and chess connectors built on MCP, the protocol that gives Claude outside tools. They solve different problems: a skill teaches Claude a way of working, while a connector hands it new tools, such as a real chess engine. This guide explains both with Anthropic's own documentation, lists what actually exists for chess, shows why the engine matters with a real position, and covers how to use a chess connector from Claude Code. Every setup command below comes from the official page we link next to it. ## What a Claude skill is Anthropic describes [Agent Skills](https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview) as folders that package instructions, metadata and optional resources such as scripts, which Claude uses automatically when they're relevant (docs checked October 26, 2026). The core is one file, `SKILL.md`. It starts with a short header holding a `name` and a `description`, then plain Markdown instructions. Three details matter for chess: 1. **Claude loads skills in stages.** At startup it only reads each skill's name and description, about 100 tokens per skill. The full instructions load when your request matches the description, and bundled files load only when needed. So the description decides when your chess skill wakes up. 2. **Skills can carry scripts.** A script runs through the command line and only its output reaches Claude. That's how a chess skill can call Stockfish: the script runs the engine and Claude reads the result. 3. **Skills live in different places on each surface.** In [Claude Code](https://code.claude.com/docs/en/skills), a personal skill goes in `~/.claude/skills//SKILL.md` and a project skill in `.claude/skills//SKILL.md`. On claude.ai you upload a zip in Settings, on paid plans with code execution turned on. Anthropic notes that custom skills don't sync between claude.ai, the API and Claude Code. One limit shapes chess skills. The overview says skills on the Claude API run with no network access, skills on claude.ai have full, partial or no network access depending on settings, and in Claude Code they have the same network access as any program on your computer. A skill that downloads your Lichess games or installs Stockfish works most naturally in Claude Code. ## What an MCP connector is The [Model Context Protocol](https://modelcontextprotocol.io/docs/getting-started/intro) is an open standard for connecting AI assistants to outside tools and data. An MCP server offers tools; the assistant decides when to call them. A remote MCP server runs on the internet, so nothing is installed on your machine; you usually sign in with OAuth, which means the service never shows Claude your password. On claude.ai, Anthropic's help article on [custom connectors using remote MCP](https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp) explains how to add one by its URL and then connect your account (checked October 2026). In Claude Code, the [MCP documentation](https://code.claude.com/docs/en/mcp) gives this syntax for a remote HTTP server: ``` claude mcp add --transport http ``` The same page says Claude Code supports OAuth 2.0 for remote servers: you run `/mcp` inside Claude Code and finish the sign-in in your browser. It also says that if you log in to Claude Code with a claude.ai account, the connectors you added in claude.ai are available in Claude Code automatically. That doesn't apply when Claude Code uses an API key or a third-party cloud provider instead. ## Claude chess skill vs chess connector Both can make Claude better at chess, but in different ways. Here is the practical difference. | | Chess skill | Chess connector (remote MCP) | |---|---|---| | What it is | A folder with `SKILL.md`, maybe scripts | A server that offers tools | | Where the engine runs | On your computer, if the skill includes one | On the provider's server | | Setup | Copy a folder, install Python packages and Stockfish | Paste a URL, sign in | | Works in | Mainly Claude Code; claude.ai with limits | Claude apps, Claude Code; some work in ChatGPT too | | Fetches your games | Only if its scripts do | If the server offers that tool | | Who keeps it updated | You, from the repository | The provider | | Main risk | Runs code on your machine | Sends data to a third-party service | A good way to think about it: a skill is a recipe, a connector is a kitchen appliance. A chess skill can tell Claude "always check a move with the engine before you explain it". It still needs an engine somewhere, either bundled in its scripts or reached through a connector. ## Claude Code chess skills you can find today We searched GitHub and the web on October 26, 2026. These are the public chess skills we could check ourselves. Star counts change. | Project | What it does | Engine | Notes | |---|---|---|---| | [brumar/chess-postmortem-skills](https://github.com/brumar/chess-postmortem-skills) | Four Claude Code skills: full game analysis, single-position study, narrated video, and playing a game against Claude | Stockfish on your computer, for analysis and positions | MIT license; 75 stars; outputs an annotated PGN and an HTML board; the author says a full run takes about an hour | | [yongqyu/claude-chess](https://github.com/yongqyu/claude-chess) | Play against Claude in the terminal, with a comment after each move and a saved game review | None: the README says no Stockfish is needed | 5 stars; the review includes an "estimated ELO" from Claude itself | The first project is the most complete Claude Code chess skill we found. Its README is honest about the trade-offs: it takes time, it suits slow games you thought hard about, and it relies on Stockfish to check what Claude writes. The second one is a fun sparring partner, but with no engine its move ratings are Claude's own judgement, with the limits described in [Can Claude play chess?](/answers/can-claude-play-chess/) Before you install either one, read Anthropic's security note in the [skills overview](https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview): use skills only from trusted sources, and audit every file, because a skill can direct Claude to run code. Open the `SKILL.md` and the scripts first. It takes ten minutes. ## Why a chess skill needs an engine: one example Language models judge positions by pattern, not by calculation. That works until a single move changes everything. Here is a position from a very common beginner trap. White's queen and bishop both aim at the pawn next to Black's king. > **Black to move. The knight move looks active but loses at once.** > Position (FEN): `r1bqkbnr/pppp1ppp/2n5/4p2Q/2B1P3/8/PPPP1PPP/RNB1K1NR b KQkq - 3 3` > Played: Nf6 — Knight to f6 attacks the queen and the e4 pawn > Better: g6 — Pawn to g6 attacks the queen and keeps the position about equal The knight move looks great: it attacks the queen and a pawn at the same time, a double attack. But it leaves the pawn next to Black's king guarded only by the king, and Stockfish finds checkmate in one: the queen takes that pawn. The engine's top choice is the pawn move instead, which attacks the queen and keeps the game about equal. A model writing from memory might praise the knight move, because "develop a knight and attack the queen" sounds like good advice. An engine never misses mate in one. That's why the better chess skills, like the post-mortem one above, use Claude to explain and Stockfish to decide what's true. For the bigger picture of how models play on their own, see our [LLM chess benchmark](/data/llm-chess-benchmark/) page. ## How to use a chess connector in Claude Code If you want engine-checked reviews without installing Stockfish, a remote connector is the shorter path. Chess Coach is one: a remote MCP server with OAuth sign-in. Based on Anthropic's [Claude Code MCP documentation](https://code.claude.com/docs/en/mcp) (checked October 26, 2026), there are two documented ways to reach a server like it. 1. **Through claude.ai.** Add it as a custom connector in claude.ai and sign in there. If you log in to Claude Code with that claude.ai account, the docs say claude.ai connectors appear in Claude Code automatically. Run `/mcp` to see them. 2. **Directly in Claude Code.** Use the documented syntax with your own name for the server and its URL, then run `/mcp` to sign in: ``` claude mcp add --transport http chess-coach https://chess-coach-246271308938.europe-west1.run.app/mcp ``` Two honest caveats. First, we built and tested Chess Coach for the Claude and ChatGPT apps, not specifically for Claude Code. Second, its interactive board is an MCP App, a page meant for the host app to draw. The Claude Code MCP docs describe these as pages for a host application to render and don't say the terminal shows them, so in Claude Code expect the review as text, not the clickable board. ## How to write your own claude code chess skill If you'd rather build your own, the format is small. Following the [Claude Code skills docs](https://code.claude.com/docs/en/skills): 1. **Create a folder** such as `~/.claude/skills/chess-review/` (personal) or `.claude/skills/chess-review/` (one project). 2. **Write `SKILL.md`** with a description that says what the skill does and when to use it, for example: "Reviews a finished chess game. Use when the user shares a PGN, a Lichess or Chess.com link, or asks why they lost." 3. **Write the instructions** in plain Markdown: which tool or script gives the facts, how to explain a mistake for the player's level, what never to do. 4. **Decide where the facts come from.** Either bundle a script that runs Stockfish on your machine, or tell Claude to call a connector's tools, for example Chess Coach's `review_game` and `show_board`. 5. **Test it** by asking for something that should match the description, and adjust the wording if Claude doesn't pick the skill up. ## Common mistakes with Claude chess skills - **Trusting a skill with no engine.** If nothing in the folder runs Stockfish or calls a connector, every evaluation is Claude's guess. - **A vague description.** "Chess helper" rarely triggers. Name the situations: a PGN, a game link, "why did I lose". - **Installing without reading.** Skills can run code. Audit them, as Anthropic advises. - **Expecting skills to follow you everywhere.** A skill in Claude Code isn't on claude.ai unless you upload it there too. - **Using it during a game.** Reviews are for finished games. Chess Coach refuses to help with games in progress, and any chess skill you install deserves the same rule. ## How Chess Coach fits in Chess Coach is not a skill. It's a connector for Claude and ChatGPT, a remote MCP server with an interactive board. It finds your finished Chess.com and Lichess games (by username, game link or pasted PGN), reviews them with Stockfish on our own server, and lets your assistant explain the key moments in plain words, with a red arrow for the move you played and a green one for a better move. You can step through moves, try your own and practise your mistakes. It never helps with games in progress. The free plan includes one in-depth review a day. [Connect it to Claude or ChatGPT](/connect/), read the [docs](/docs.html), or compare it with a local engine in [Stockfish MCP server](/answers/stockfish-mcp-server/). ## FAQ ### Is there an official Claude chess skill? No. As of October 26, 2026, Anthropic's pre-built skills cover PowerPoint, Excel, Word and PDF, and its open-source [skills repository](https://github.com/anthropics/skills) doesn't include a chess skill. The chess skills you can find are community projects on GitHub. ### What is the difference between a Claude skill and an MCP connector? A skill is a folder of instructions (a SKILL.md file, plus optional scripts) that Claude reads when a task matches. An MCP connector is a server that gives Claude tools to call, often a remote one you sign in to. A chess skill tells Claude how to work; a chess connector gives it an engine and your games. ### Can I use Claude Code for chess? Yes. Claude Code supports custom skills in ~/.claude/skills/ or .claude/skills/, and, according to Anthropic's docs, remote HTTP MCP servers with OAuth sign-in. Community chess skills run Stockfish on your own computer; a remote connector runs the engine on its server. ### Is Chess Coach a Claude skill? No. Chess Coach is a remote MCP connector for Claude and ChatGPT. It reviews your finished Chess.com and Lichess games with Stockfish on our server and shows them on an interactive board in the Claude and ChatGPT apps. ### Are chess skills from GitHub safe to install? Treat them like any software you download. Anthropic advises using skills only from trusted sources and auditing every file, because a skill can make Claude run code. Read the SKILL.md and its scripts before you copy the folder. ## Related - [Chess MCP server for Claude and ChatGPT: Chess Coach](https://thechesscoach.netlify.app/connect.md) - [Stockfish MCP server: give Claude a chess engine](https://thechesscoach.netlify.app/answers/stockfish-mcp-server.md) - [Can Claude analyze chess games? Yes, with Stockfish](https://thechesscoach.netlify.app/answers/can-claude-analyze-chess-games.md) - [Can you play chess with Claude? How it works (2026)](https://thechesscoach.netlify.app/guides/play-chess-with-claude.md) - [LLM chess benchmark: 4 public leaderboards (Oct 2026)](https://thechesscoach.netlify.app/data/llm-chess-benchmark.md)