What it is
Claude Code is Anthropic’s AI coding agent. You run it in a terminal inside a project, describe what you want in plain language, and it reads the code, edits files, runs commands and tests, and explains what it did. It asks before it does anything you haven’t allowed.
It works with Git, understands large codebases, and can be extended with your own commands, hooks and MCP servers.
Why it’s here
- Works where the code is. It runs in your terminal and project directory, next to your editor and tools.
- Reads before it writes. It looks through the codebase to understand it, then makes focused changes you can review.
- Contained. Every command it runs goes through its own sandbox, limited to your project and an allow list of sites.
On CoreDesktop
CoreDesktop installs Claude Code with Anthropic’s official native installer, which puts claude in ~/.local/bin and keeps itself up to date; CoreDesktop adds ~/.local/bin to your PATH. If you don’t already have ~/.claude/settings.json, it writes one with hooks that show a desktop notification when Claude needs your input and when it finishes a task.
The other agents run in CoreDesktop’s sandbox, but Claude Code can’t, because bubblewrap can’t run inside itself. Instead CoreDesktop turns on Claude Code’s own built-in sandbox and gives it the same list of allowed sites; anything you’ve already set in the sandbox section of your settings wins. Type /sandbox inside Claude to see it. The other agents can’t read Claude’s login. See AI sandbox for how the sandboxes work.
Before the first use, run configure claude and choose how you pay: a Claude subscription (Pro, Max, Team or Enterprise) or the Anthropic Console. Then start it from a project directory with claude.