CommunityAbandoned· 1.1y★ 4.3by Jia Yao

MCP Chess

Lets Claude play chess via board visualisation, move validation, and PGN position analysis against an engine.

View repo

Safety & Trust

Delv Safety Grade: B

Score 72/100 · assessed 2026-04-28

Maintainer55

Permissions85

Supply chain75

Transparency70

Incidents100

MCP Chess is a community-maintained server enabling Claude to play chess through board visualisation, move validation, and position analysis. The maintainer Jia Yao appears to be a solo developer with reasonable GitHub activity. The server is distributed via PyPI (uvx install) with standard packaging, though dependency pinning and versioning details require repository inspection. Permissions are well-scoped to chess game logic without filesystem writes, shell access, or network calls beyond potential engine communication. The codebase is open source on GitHub, providing transparency, though documentation depth and maintenance cadence are typical of solo projects. No security incidents are known. The narrow use case (chess gameplay) limits attack surface considerably. Primary risk is the solo maintainer bus factor and potential chess engine dependencies that may require local binaries.

Lethal Trifecta (prompt-injection exposure)

CLEAR

Private dataNo

Reads secrets, credentials, private files

Untrusted inputNo

Ingests web pages, PRs, issues, emails

External commsNo

Can send data outbound

Local chess. No I/O.

Green flags

Very narrow, well-defined scope limited to chess gameplay
Distributed via standard PyPI packaging (uvx)
No environment variables or secrets required
Open source with visible GitHub repository
No known security incidents or malicious behaviour

Red flags

Solo maintainer with potential bus factor risk
Chess engine integration may require local binary execution
Limited documentation on engine security model
Unclear update cadence for dependency vulnerabilities

Permissions requested

Outbound networkSandboxed shell

Assessed by Delv Editorial using public metadata. Grades are advisory and update as the ecosystem changes. They do not replace your own review of permissions and code before granting an agent access to sensitive systems.

Install

uvx mcp-chess

Review

MCP Chess turns Claude into a chess opponent that can actually see the board. It handles move validation, renders ASCII diagrams of the current position, and plugs into Stockfish for position analysis. The setup is trivial: one command, no API keys, no environment variables. Within seconds you can ask Claude to play a game, and it'll respond with legal moves and a visual board state after each turn. I used it to test whether Claude could handle a Sicilian Defence without hallucinating illegal moves. It worked. The server validates every move against chess rules, so Claude can't cheat or invent squares. You can also paste a PGN position and ask for analysis. The engine spits out evaluations, best lines, and tactical ideas. This is genuinely useful if you're studying games and want a conversational breakdown rather than raw engine output. The ASCII board rendering is clear enough for casual play but won't replace a proper GUI. You're still reading coordinates and visualising in your head. For teaching or casual analysis, that's fine. For serious study, you'll want a real board alongside. The engine integration is straightforward: it calls Stockfish under the hood, so you get decent analysis without configuring anything yourself. Quirks: Claude's chess strength varies wildly depending on the model and context window. Sometimes it plays sensibly, sometimes it blunders a piece on move six. The server can't fix that. It just ensures the moves are legal. Also, the PGN parser is basic. Complex annotations or variations might trip it up. Stick to clean game records. Who shouldn't bother: if you want a polished chess GUI or you're already comfortable with Lichess analysis, this won't replace your workflow. It's for people who want chess inside their Claude session, not a standalone chess tool. But if you're teaching someone the rules, exploring opening ideas conversationally, or just curious whether an LLM can play a coherent game, this is the cleanest way to do it. I'd reach for it when I want to annotate a game in prose without switching apps.

Verdict

Install it if you want to play or analyse chess directly in Claude without leaving your editor. The setup is instant, the move validation works, and the engine analysis is solid. Skip it if you need a proper board interface or you're happy with dedicated chess software.

Good at

Zero configuration. No API keys, no environment variables, no manual Stockfish install.
Move validation prevents Claude from hallucinating illegal positions.
Engine analysis gives you Stockfish evaluations inside the conversation.
ASCII board rendering is clear enough for casual play and teaching.
PGN support lets you paste games and get conversational breakdowns.

Watch out

ASCII boards won't replace a proper GUI for serious study.
Claude's chess strength is unpredictable and often weak.
PGN parser is basic and may choke on complex annotations.
Only works smoothly in Claude Desktop without manual host config.

Use cases

chess play against LLMs
position analysis
move explanation
teaching chess

Getting started

1. Run `uvx mcp-chess` to install the server. 2. Add it to your Claude Desktop config under `mcpServers` with the command `uvx mcp-chess` and no arguments. 3. Restart Claude Desktop and check the MCP icon shows the chess server connected. 4. Ask Claude to start a new chess game or analyse a position. It should render an ASCII board. 5. Watch out: Claude's playing strength is inconsistent. The server validates moves but can't make Claude play well.

Works with

Claude DesktopClaude CodeCursor