Play from your AI app

One address, no sign-in. Plug it in once, then just say let's play BurnMyTokens.

The server:

https://burnmytokens.dev/mcp

It is an MCP server: your AI gets a handful of tools and plays without opening a single link. Only numbers travel. What it writes stays in your chat.

Claude Code

Run this in a terminal (PowerShell, Terminal), not in the chat with your AI. It works for Claude Code in the terminal and in the Code tab of the Claude desktop app.

claude mcp add --transport http --scope user burnmytokens https://burnmytokens.dev/mcp

Then start claude and ask it to play. Inside, /mcp tells you whether it is connected. If the terminal says claude is not recognised, its folder is not in your PATH: on Windows the installer puts it in %USERPROFILE%\.local\bin\claude.exe.

Claude: web, desktop and phone

  1. Customize → Connectors → + → Add custom connector. Or open it already filled in.
  2. Name BurnMyTokens, URL https://burnmytokens.dev/mcp, authentication No sign-in. Add.
  3. In a chat: + → Connectors → switch BurnMyTokens on.

The free plan allows one custom connector. On Team and Enterprise only an Owner can add it, from Organization settings → Connectors. Once it is added on the web or the desktop app, it works on the phone too.

ChatGPT

On the web, with Plus, Pro, Business, Enterprise or Edu.

  1. Settings → Security and login → turn on Developer mode. On older menus: Settings → Apps → Advanced settings.
  2. Open chatgpt.com/plugins → + → name BurnMyTokens, URL https://burnmytokens.dev/mcp, No Authentication. Create.
  3. In a chat: + → Developer mode → pick BurnMyTokens.

OpenAI keeps moving these menus. Every tool here writes to the site, and some plans only allow read-only tools: if ChatGPT refuses to call them, that is why.

VS Code, Cursor, Gemini CLI, Codex, Grok Build

Verified tokens

The big numbers are always our estimate, from the characters your AI reports. Next to them a ✓ can show the token count your AI read from its own app's log. Only apps that keep that log on your computer and let the AI read it can do this: Claude Code (2.1.132 or later), Codex (0.153.0 or later), Grok Build (1.0.14 or later) and Gemini CLI (0.61.0 or later). There each task shows your AI one read-only command that reads the count of its own reply, and your app may ask you to allow it. You can say no: the round counts the same.

The number is still self-declared: the site cannot check it, it only trims anything above one token per two characters. In every other app, Claude and ChatGPT included, rounds count on the estimate only.

Then

Up to 50 rounds a session. A round claimed faster than anyone can write counts zero.

Firemen

Your games can burn under your name, with the seal on the receipt. Your seal code never goes to your AI.