Ask HN:How AI/MCPs can be useful?

I have been writing commercial code for over a decade an been exposed to things like coding, architecture, devops, team management, etc.

I've recently tried that AI vscode alternative (not windsurf, another one popular, forgot the name).

Sometimes code completion is a-okay, most of the times it'd produce a lot of code that I'd spend more time understanding than actually writing it. Sometimes when I expect completion to occur, it just doesn't. And I mostly rely on snippets and vscode auto-completion anyways.

It's a-okay for writing repetitive test code, however, I also noticed that if I set-up my framework well, then I can copy/paste most of the stuff and AI auto-completion does not add a lot of value.

Granted, I'm using claude as a search engine – it's pretty decent as a standalone app and gives good information to work with. It helps me understand unfamiliar APIs/frameworks/concepts quicker. Anyone takes away claude, I can live with RTFM.

I've recently tried connecting MCPs with claude desktop. It's cute that it can fetch stuff, or populate a local db or whatever. These seems to be gimmicks.

My question is – how can I really make MCPs and AI more useful for software development / team management processes? Can you share stories of how you use AI to really improve workflows?