Are you actually using MCP?

Honestly, I barely use it for development. I’ve tried a few. None felt essential.

The reason is simple — development is mostly non-routine work.

When I throw things at Claude Code like

it usually figures it out through the shell.

So at first, I had it completely wrong.

“Claude Code must be running with MCP built in from the start.”

That’s what I thought.

It connects to DBs, hits external APIs, reads files. So I figured “MCP is great.”

Then later I noticed.

Wait — that wasn’t MCP.

Claude was just using the CLI and shell like normal.

Someone has to build the MCP server

When you think about it, MCP requires

“someone to build and publish an MCP server.”

Things like

only exist if the official vendor or a third party has actually shipped one. Without that, it doesn’t work.

So from a developer’s perspective, the conclusion tends to be

“why not just call the CLI?”

That’s especially true for infra and backend.

AWS CLI kubectl docker psql curl

These are too strong.

So is MCP useless? Not really.

It probably shines in “business operations.” Things like

Automating this kind of “human office work.”

That’s where MCP really fits.

Development, on the other hand, has too much freedom.

For developers, it tends to land at

“writing bash on the spot is faster.”

So far, the one case where I’ve thought “MCP is the better tool” is Playwright.

Browser operation gets handed to AI in a structured way, so testing and visual checks fit really well.

Other than that, my current sense is

“the CLI is enough.”

So lately my mental model is:

MCP = a business-integration protocol for the AI era.

But developers come from a CLI culture, so it doesn’t feel revolutionary.

If anything, it feels more like a “revolution for non-engineers.”