Using Claude in VS Code: what Claude Code actually does

What Claude Code is

Most people meet Claude in a chat window. Claude Code is the other way to use it: a coding agent that runs in your terminal, and as an extension inside VS Code and JetBrains editors. Instead of copying code back and forth from a chat, it works directly on the project open in your editor, reading the files, making edits, and running commands.

The distinction that trips people up: this isn't smarter autocomplete. Autocomplete finishes the line you're typing. Claude Code takes a goal ("add a login page", "fix this failing test", "turn this CSV into a chart") and works through it, file by file, the way a developer would, while you watch and approve.

How it works in VS Code

Once the extension is installed, Claude Code sits alongside your code. A normal session looks like this:

• You describe what you want in plain English. No special syntax.
• It plans the work and, on anything consequential, shows you the plan before acting.
• It reads the relevant files itself, so you don't have to paste them in. It already has the context of your project.
• It edits files, runs the tests or the app, reads the output, and fixes what broke, looping until the task is done.
• You approve the steps that touch real things, committing code, running a command, so nothing happens behind your back.

That approve-as-it-goes rhythm is the whole point. The agent does the legwork; you stay the decision-maker. It's why people who aren't full-time engineers can still get real results with it, and why engineers trust it on actual codebases.

MCP: how it reaches your other tools

On its own, a coding agent can read and write files. The thing that makes it genuinely useful for business software is the Model Context Protocol, or MCP, an open standard for connecting Claude to your other systems: a database, an internal API, your issue tracker, a design file, a spreadsheet.

With the right MCP connection in place, "build a page that lists our customers" can actually query your real customer data instead of inventing a fake table. MCP is the difference between a clever demo and a tool that does your real work. It's also where a little training pays off, because knowing which tools to connect, and which to keep it away from, is a skill of its own.

Getting started, and getting good

Starting is genuinely quick: install the VS Code extension, open a project, and ask it to do something small and real, fix a bug, add a field, explain a file you don't understand. You'll learn more from one real task than from any amount of reading.

Getting good is a different curve, and it's mostly about judgement: how to describe a task so the agent gets it right the first time, when to let it run and when to step in, and how to keep it safe around your data. That judgement is exactly what we drill in our hands-on AI training. And if you just want to see it in motion, we build something live with Claude every Saturday in AI Pulse, come watch how an experienced builder actually drives it.