LABSCO SUMMARY

A thin, honest UCI bridge — the strength is Stockfish's.

Because it passes UCI through rather than inventing an abstraction, anything the engine understands works, and the choice between searching to a depth and thinking for a fixed number of milliseconds stays yours. The session model is what makes it usable in a conversation: the position persists between calls instead of being resent each time.

What it is

A Go server that bridges an assistant to Stockfish over the UCI protocol, running several concurrent engine sessions and returning every answer as JSON.

What you get

UCI passed straight through — uci, isready, position startpos, position fen [FEN], go, go depth [n], go movetime [ms], stop, quit
A command parameter and an optional session_id; leave the session out and one is created for you
Concurrent sessions, each keeping its own UCI state between calls, so several positions can be analysed side by side
Structured responses: status, session id, the command sent, the engine's reply lines, and an error field when something went wrong
Sessions cleaned up on quit or on timeout, with a cap so a runaway loop cannot fork-bomb the machine

Requirements

A Stockfish binary on the machine — MCP_STOCKFISH_PATH points at it and defaults to stockfish on PATH. Clone the repo and make install to build the server. stdio is the default mode; MCP_STOCKFISH_SERVER_MODE=http switches to HTTP with host and port variables. MCP_STOCKFISH_MAX_SESSIONS defaults to 10, session timeout to 30m, command timeout to 30s.

Setup effort

One command — mcp-stockfish

See installation options

USE CASES

01

Getting the engine's view of a position

Set the position from FEN, then ask for the best move at the depth you want.

02

Analysing several games at once

Each session keeps its own board, so positions don't overwrite one another.

03

Keeping analysis inside a time budget

Ask it to think for a fixed number of milliseconds rather than to a fixed depth.

04

Stopping a search that is taking too long

Send the stop command against that session and the engine returns what it has.

BEFORE YOU INSTALL

What to know first

Requires

mcp-stockfish on PATH

Where it runs

The documented client runs the server on your machine.

INSTALLATION

Connect your client

Server setup

TERMINAL

mcp-stockfish

MCP Clients

Claude Desktop

CLAUDE DESKTOP · JSON

{
  "mcpServers": {
    "chess": {
      "command": "mcp-stockfish",
      "env": {
        "MCP_STOCKFISH_LOG_LEVEL": "info"
      }
    }
  }
}

Stockfish