This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
A Model Context Protocol (MCP) server providing AI assistants with professional-grade chess analysis using Stockfish (alpha-beta) and optionally Leela Chess Zero / Lc0 (neural network). Both engines share the same tool interface and run side-by-side when Lc0 is configured.
Tool | Description |
| Analyse any position (FEN → evaluation + best moves + top lines) |
| Full game analysis (PGN → move-by-move eval, accuracy %, error counts) |
| Search opening database by name or ECO code |
| Identify the opening from moves or PGN |
| Generate tactic puzzles from positions |
LC0_WEIGHTS_PATH is set)Tool | Description |
| Analyse a position with the Lc0 neural network |
| Full game analysis using Lc0 evaluation |
| Generate tactic puzzles using Lc0's evaluation |
Related MCP server: Chess MCP Server
# Published image (Stockfish + Lc0 + Maia-1900 baked in; linux/amd64 + arm64)
docker run -i ghcr.io/alegerber/stockfish-lc0-mcp:latest
# …or build it yourself — Stockfish only
docker build -t stockfish-lc0-mcp .
docker run -i stockfish-lc0-mcp
# With Lc0 (mount weights file)
docker run -i \
-e LC0_WEIGHTS_PATH=/weights/lc0.pb.gz \
-v /path/to/weights:/weights \
stockfish-lc0-mcp
# Or with docker compose
docker compose up --buildPrerequisites: Node.js 22+ and a locally installed Stockfish — see the install commands under Option 3. The npm package does not bundle the engines — the Docker image does.
npx stockfish-lc0-mcpPrerequisites: Node.js 22+ (CI-tested on LTS 22, 24, 26), Stockfish binary installed.
# Install Stockfish
# macOS: brew install stockfish
# Ubuntu: sudo apt install stockfish
# Windows: download from https://stockfishchess.org/download/
# Install dependencies and build
npm install
npm run build
# Run (Stockfish only)
npm start
# Run with Lc0
LC0_WEIGHTS_PATH=/path/to/lc0.pb.gz npm startVariable | Default | Description |
|
| Path to the Stockfish binary |
|
| Number of CPU threads |
|
| Hash table size in MB |
Variable | Default | Description |
| (unset) | Path to a |
|
| Path to the Lc0 binary |
| (auto) | Lc0 backend: |
|
| Number of CPU threads for Lc0 |
|
| Hash table size in MB for Lc0 |
Note: The
depthparameter for Lc0 tools is mapped internally to node counts via an exponential table (100 nodes at depth 1 → 1 000 000 nodes at depth 30), since MCTS depth is not comparable to alpha-beta depth.
Add to your claude_desktop_config.json:
{
"mcpServers": {
"chess": {
"command": "docker",
"args": ["run", "-i", "--rm", "stockfish-lc0-mcp"]
}
}
}{
"mcpServers": {
"chess": {
"command": "docker",
"args": [
"run", "-i", "--rm",
"-e", "LC0_WEIGHTS_PATH=/weights/lc0.pb.gz",
"-v", "/path/to/weights:/weights",
"stockfish-lc0-mcp"
]
}
}
}Requires a locally installed Stockfish (see Quick Start). Set STOCKFISH_PATH explicitly: GUI-launched clients (e.g. Claude Desktop on macOS) don't inherit your shell's PATH, so a bare stockfish lookup can fail even though brew install stockfish succeeded.
{
"mcpServers": {
"chess": {
"command": "npx",
"args": ["-y", "stockfish-lc0-mcp"],
"env": {
"STOCKFISH_PATH": "/opt/homebrew/bin/stockfish"
}
}
}
}{
"mcpServers": {
"chess": {
"command": "node",
"args": ["/path/to/stockfish-lc0-mcp/dist/index.js"],
"env": {
"STOCKFISH_PATH": "stockfish",
"STOCKFISH_THREADS": "2",
"STOCKFISH_HASH": "256",
"LC0_PATH": "lc0",
"LC0_WEIGHTS_PATH": "/path/to/lc0.pb.gz"
}
}
}
}"Analyse this position: rnbqkbnr/pppppppp/8/8/4P3/8/PPPP1PPP/RNBQKBNR b KQkq - 0 1"
"Review this game: 1. e4 e5 2. Qh5 Nc6 3. Nf3 g6 4. Qh4 Be7 ..."
"What is the Wayward Queen Attack?"
"What opening is 1. e4 e5 2. Nf3 Nc6 3. Bc4?"
"Create a tactic puzzle from this position: [FEN]"
"What does the neural network think of this position?"
"Analyse this game with Lc0 and compare with Stockfish: 1. e4 e5 ..."
"Find tactics using Lc0 in this position: [FEN]"
src/
├── index.ts # MCP server entry, tool registration (Stockfish + Lc0)
├── types.ts # TypeScript interfaces (UciEngine, UciLine, UciScore, …)
├── constants.ts # Thresholds, defaults, LC0_DEPTH_TO_NODES mapping
├── schemas/
│ └── index.ts # Zod input validation schemas
├── services/
│ ├── engine.ts # BaseUciEngine, StockfishEngine, Lc0Engine
│ ├── chess-utils.ts # chess.js wrapper (PGN/FEN/SAN/openings)
│ └── formatting.ts # Markdown output formatting
└── tools/
├── analyse-position.ts # Single position analysis
├── analyse-game.ts # Full game analysis + accuracy model
├── openings.ts # Opening lookup & identification
└── puzzle.ts # Tactic puzzle generationBoth engines implement the UciEngine interface and are interchangeable at the tool layer — all tool functions accept a UciEngine parameter, so sf_* and lc0_* tools share identical logic with different engine instances.
npm install # Install dependencies
npm run build # Compile TypeScript → dist/
npm test # Run unit tests (Vitest, 150+ tests)
npm run lint # ESLintMIT
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Chess MCP for Claude: engine analysis, attack maps, game review. One URL, no install.
Chess tools: game reviews for Chess.com and Lichess, FIDE ratings, openings, positions, puzzles.
Review your Chess.com and Lichess games with Stockfish on an interactive board, in plain words.
Pedagogical chess intelligence for AI agents: explain positions and games for a target Elo.