The pitch is narrow and honest: language models draw numbers unevenly, and this replaces that draw with one from random.org's atmospheric noise. There is no configuration and no key, but there is a build step and a network dependency, so it belongs in setups where the fairness of the draw actually matters.
An MCP server that fetches true random numbers from random.org, which derives them from atmospheric noise. The repository opens with a distribution chart of 100 numbers in the range 1-100 drawn from three popular LLMs, to show the bias it exists to remove.
- Random numbers from an external physical source rather than from the model's own token distribution
- A stdio server that plugs into any MCP client through a plain command entry
No account and no key. Clone the repository, run `npm install` and `npm run build`, then point your client at the absolute path of `build/index.js` with `node`. The package is `mcp-random-number` (1.0.0 in package.json).