𝗧𝗵𝗲 𝗗𝗮𝘆 𝗔𝗜 𝗔𝗿𝗴𝘂𝗲𝗱 𝗪𝗶𝘁𝗵 𝗠𝗗𝗡 𝗔𝗻𝗱 𝗟𝗼𝘀𝘁
AI coding assistants change how you write software. They write tests and explain APIs. But they often fail at one thing: knowing what browsers support today.
Mozilla recently tested this. They used Claude Code to ask about new Firefox features. The AI confidently claimed Firefox did not support the Web Serial API. It even said Mozilla had no plans to add it.
The AI was wrong. Firefox 151 had already shipped that feature.
This happens because the web moves faster than AI training data. Browsers add new CSS and HTML features every few weeks. An AI trained months ago will guess when it lacks current data. It produces answers that look correct but are outdated.
Mozilla solved this with the MDN MCP Server.
MCP stands for Model Context Protocol. It lets AI models connect to external data. Instead of relying on memory, the AI checks the source.
Old way: Developer → AI Memory → Answer
New way: Developer → AI → MDN → Answer
The MDN MCP Server gives the AI direct access to documentation and browser compatibility data. Mozilla tested this and found three big improvements:
- Compatibility data became accurate.
- The AI correctly identified new features.
- The AI stopped making unsupported assumptions.
Surprisingly, the AI became twice as fast. Without MCP, the assistant has to search the web and parse HTML pages. With MCP, it gets structured data immediately. It spends less time searching and more time answering.
You can use this with many tools like Claude Desktop, Cursor, or VS Code.
To add it to Claude Code, run this command: claude mcp add --transport http mdn https://mcp.mdn.mozilla.net/
The lesson is simple. We should not treat AI as a replacement for documentation. The future is AI that reads the documentation with you.
Source: https://dev.to/josephciullo/the-day-ai-argued-with-mdn-and-lost-mm7