My AI assistant kept guessing about data breaches. So I gave it one free URL (MCP, no API key)

When asking an AI assistant if your email has been in a data breach, you receive a confident answer, but it's just a guess. The AI model cannot look anything up, so it relies on old training data to fill the gap. However, a guess is not a reliable answer when it comes to your own account. To address this issue, the author created XposedOrNot, a free and open-source breach index.

The author initially copied API results into chat windows by hand, but this felt tedious. So, they added an MCP (Multi-Modal Communication Protocol) endpoint to the API. MCP is a standard way to give an AI assistant tools it can call. By pointing the AI at a server, it determines when to use what tool.

Setting up the MCP was simple, taking about two minutes. The author provided instructions for various platforms, such as using Claude Code, Cursor, and VS Code. Once the setup was complete, the AI could read the tool list and pick one to answer questions like "has my email been in any breaches, and what was leaked?" The assistant can use six different tools, each serving a specific purpose related to checking email breaches, analyzing breach data, and managing the breach index.

Through this process, the author learned several important lessons. First, keep the system stateless, as each request should stand on its own. Second, inform the client that the tools are read-only and idempotent, minimizing the need for permission requests. Third, provide clear error messages for the model, allowing it to fix any issues and try again. Lastly, limit the free version of the service, encouraging users to upgrade to a paid plan if they need more functionality.