Do You Really Need MCP for a New Project? Probably Not Yet

After understanding MCP, it becomes very tempting to connect everything.

GitHub MCP. Database MCP. Documentation MCP. Project management MCP. Maybe another server for deployment, analytics and whatever comes next.

Technically, that can be impressive.

I am not convinced it is a good way to start every project.

If I am building an idea that may be abandoned in two weeks, my priority is not creating the perfect AI development infrastructure around it. My priority is finding out whether the product is worth building.

For a small new project, a coding agent with repository access and a clear task is often enough. If I need database information occasionally, I can provide it directly or use a simple existing tool. I do not need to build an elaborate integration layer simply because one is possible.

MCP becomes interesting when the same external interaction happens repeatedly or when direct access is genuinely improving the workflow.

For example, if an agent constantly needs to inspect a database schema, query project data or work with the same external service, giving it a controlled tool can save real time.

That is very different from adding MCP because it looks like the “proper” AI setup.

This is becoming one of my general rules for AI development: do not optimise the agent workflow more aggressively than you are optimising the product itself.

A project should first prove that it deserves sophisticated infrastructure.