Play from your AI agent
Ludobench has an MCP server. MCP, the Model Context Protocol, is how AI agents use other services. Connect it to your agent — Claude, ChatGPT, Cursor, VS Code or another — and ask it for a game: it chooses the game and its options, makes the table in your account and gives you the link. You open the link and play.
The agent never plays for you. It cannot see a game in progress or chat; it makes tables, lists yours and calls them off, and nothing else.
The server's address
https://mcp.ludobench.com/mcpWhen you connect it, your agent opens a Ludobench page in your browser: sign in there and allow it. You need a free account.
Things to ask
- Make me a game of backgammon against the Practice Bot, to five points.
- Set up a blitz chess table for me and a friend, and give me the invite link.
- Start a hard Sudoku.
- Which card games can four people play together?
Connecting it
Claude (claude.ai and the desktop app)
- Open Settings → Connectors.
- Choose Add custom connector, name it Ludobench and paste the address.
- Choose Connect, and allow it on the Ludobench page that opens.
On the Free plan Claude takes one custom connector.
Claude Code
- Run this command, then /mcp in Claude Code: choose ludobench and sign in.
claude mcp add --transport http ludobench https://mcp.ludobench.com/mcpChatGPT
- Turn on developer mode in ChatGPT's settings (under Apps or Connectors, then Advanced).
- Create a connector with the address, and OAuth as its sign-in.
- Allow it on the Ludobench page that opens.
Developer mode is not on every plan, and the menus' names change often.
Cursor
- Put this in ~/.cursor/mcp.json, then sign in when Cursor asks.
{ "mcpServers": { "ludobench": { "url": "https://mcp.ludobench.com/mcp" } } }VS Code
- Put this in .vscode/mcp.json (or your profile's mcp.json), start the server and sign in.
{ "servers": { "ludobench": { "type": "http", "url": "https://mcp.ludobench.com/mcp" } } }Any other agent
- An agent that cannot sign in by itself can use a personal token: make one in Settings → AI agents, and have the agent send it with every request, as this header:
Authorization: Bearer lb_pat_…What your agent can do
- list_games
- lists the games, and how each is played.
- describe_game
- tells one game's lengths, clocks and options, and which AI models you may play against.
- create_table
- makes a table: against the computer (the free Practice Bots, or the AI models with Plus), with friends (an invite link and a four-digit room code), open to anyone, or alone — at the length, clock and options you ask for.
- my_tables
- lists your waiting tables and your live match.
- cancel_table
- calls off a table that has not started.
How the tables work
- A table against the computer, or a game alone, waits for you: it starts when you open its link, so no clock runs while you read your agent's answer.
- Everything else is as in the game's lobby: one live match at a time, the AI models with Plus, and games against the computer are unrated.
- An agent makes at most 30 tables an hour.
Your control
Every app you allow and every token you make is listed in Settings → AI agents, with when it was last used. Revoke one and it stops at once. The agent sees nothing of your account but your tables.