What Is Claude Code?

Claude Code is Anthropic's agentic coding tool: a program you run in a terminal that gives Claude direct, persistent access to a real development environment — your files, your shell, your git history — instead of a text box that only ever sees what you paste into it.

The core idea

A chat assistant answers a question and stops. Claude Code runs a loop: it reads what it needs, decides on an action, takes that action through a tool (reading a file, editing one, running a command), looks at the result, and decides what to do next — repeating until the task is done or it genuinely needs your input. That loop is the entire mechanism behind everything else in this guide: plan mode, subagents, hooks, and permission modes are all ways of shaping how that loop behaves, not separate features bolted on top of it.

This matters in practice because it changes what kind of requests are worth making. "Explain what this function does" works fine in a chat window. "Find every place this function is called, update the call sites for the new signature, and confirm the test suite still passes" needs an agent that can actually go look, actually make the edits, and actually run the tests — which is the difference between a chat assistant and Claude Code.

What it's built on, and what's added on top

Claude Code uses the same underlying Claude models available through the API and claude.ai. What Claude Code adds is the agent loop itself, a fixed and well-tested set of tools for software engineering (file operations, shell execution, search, git, web access), and the surrounding product layer that makes that loop usable day to day: permission controls so it doesn't run unattended actions you haven't agreed to, project memory so you don't re-explain your codebase every session, and configuration surfaces (hooks, custom commands, skills, MCP servers) for shaping its behavior to a specific project or team.

You could build something similar yourself against the API with tool use and a loop of your own — several later topics in this guide describe mechanics, like subagents and background tasks, that mirror patterns available to anyone building on the API. The difference is that Claude Code ships all of it together, tuned specifically for the day-to-day work of reading, writing, and running code.

What it's actually used for

In practice, the common cases are: implementing a feature end to end, debugging a failure by reproducing it and fixing the root cause, refactoring across many files consistently, writing tests for existing code, reviewing a diff before it ships, and answering questions about an unfamiliar codebase by actually reading it rather than guessing. None of these require you to write a single line of glue code — you describe the task, and the agent loop handles the rest, subject to whatever permission boundaries you've set.

The next topic covers getting it installed and running on a real project.