The terminal CLI is the main way to run Claude Code. You install it once, then start a session with a single command in any project directory.

What it does

The claude command launches an interactive session wired to your repo. It reads CLAUDE.md, loads your skills and MCP servers, respects permission settings, and runs tools like Read, Edit, and Bash. Output streams to your terminal in real time, and you can swap models or modes mid-session without restarting.

When to use it

You want the fastest path into Claude Code on a new project.
You prefer a terminal-first workflow over an IDE extension.
You need full access to tools, hooks, skills, and MCP without UI overhead.
You're scripting and want the same binary that powers every other surface.

Gotchas

First run prompts for auth. Don't commit the token that gets cached in ~/.claude/.
Running claude outside a git repo still works but skips repo-aware features like worktrees and PR status.
If you launch inside a tmux or screen session, set your terminal type correctly or keybindings may misbehave.

Official docs: https://code.claude.com/docs/en/quickstart.md