HN turns on MCP

A technical critique arguing that MCP was always a bad idea is getting real attention on Hacker News, with 311 points and 291 comments. That does not prove the argument is right, but it does tell you the developer community is actively questioning a pattern that has been gaining adoption. You can read the original post and judge the case for yourself.

For a business owner, the practical point is not to panic or rewrite anything. It’s to avoid treating a popular AI integration approach as settled infrastructure too early. When tooling proliferates quickly, teams often build around what is easiest to connect today, then discover later that reliability, security, maintenance, or control are harder than expected. The details of this critique matter less than the fact that credible technical pushback is now attracting sustained discussion.

If your team is using MCP, or planning around it, ask a simple question: what happens if the preferred approach changes in six months? Keep your AI workflows modular where you can, document the dependencies, and avoid tying critical processes to one emerging standard without a fallback. This is the kind of thing we build into an AI command centre, visibility over what your AI systems rely on and where the operational risk sits.