Cook with an assistant

Connect Claude, ChatGPT, Cursor or any MCP client to the book, then ask it to find a recipe, scale it, work from your fridge or write the shopping list.

1. Get a key

The server needs a key, one per person, so it stays quick for the people it is for. Keys are handed out by Thomas Chaplin: ask for one, and it arrives looking like ck_….

Wherever YOUR_KEY appears below, put your key in its place.

2. Connect your assistant

Nothing to install: the book is a remote MCP server (Model Context Protocol), and every assistant that speaks MCP can be pointed at it.

Claude — on the web, desktop and phone

In Settings → Connectors, choose Add custom connector, call it Chaplin Kitchen and paste this address, key included. Leave the advanced OAuth settings empty: the key in the address is all it needs. Once added on the web or desktop, it is there on the phone too.

https://chaplinkitchen.com/mcp/?key=YOUR_KEY

Claude Code

claude mcp add --transport http chaplin-kitchen https://chaplinkitchen.com/mcp/ \
  --header "Authorization: Bearer YOUR_KEY"

ChatGPT

With developer mode switched on (under Settings → Apps → Advanced), create an app with the same address as Claude, key included, and No authentication — again, the key in the address is the authentication.

https://chaplinkitchen.com/mcp/?key=YOUR_KEY

Cursor, VS Code and the rest

Add a remote (Streamable HTTP) server. Where the client lets you set headers, send the key as a bearer token — here in Cursor’s mcp.json:

{
  "mcpServers": {
    "chaplin-kitchen": {
      "url": "https://chaplinkitchen.com/mcp/",
      "headers": {
        "Authorization": "Bearer YOUR_KEY"
      }
    }
  }
}

Where it takes nothing but an address, use the one with ?key= on the end.

3. Ask

Talk to it as you would to someone who has the book open in front of them. It uses the site’s own search, scaler and fridge matching, so it gives the answers the site would.

  • “Find me something with chicken and orzo.”
  • “Give me the pancake recipe for six people.”
  • “I have eggs, spinach and feta — what can I make?”
  • “What should I cook tonight?”
  • “Write one shopping list for the dhal and the American pancakes, the dhal doubled.”
  • “Anything Italian that takes under 30 minutes?”

What it will and won’t say

  • Every recipe it gives you comes from the book, with a link to its page here.
  • Quantities are metric with British spoons, and a scaled recipe says what was scaled and what was not — the times stay as written.
  • It is told not to make up nutrition figures or cooking times, because no recipe here has any it has not stated.
  • It reads the book and changes nothing in it.

Your key

Keep it to yourself: anyone with your key, or the address with your key in it, uses the book as you. Each key has a limit on how many requests it can make a minute — more than enough to cook with. A busy assistant that goes over it is asked to wait a moment. A key that turns up somewhere it shouldn’t can be switched off, and a new one sent.

No key?

An assistant that can read the web can still read the book. llms.txt lists every recipe, llms-full.txt is the whole book in one file, and every recipe has a plain Markdown copy at its own address with .md on the end — porridge.md, say. Give your assistant one of those and ask away.