Claude Code

Claude Code is Anthropic's official CLI for Claude. It runs in your terminal, reads your codebase, executes commands, and can handle multi-step programming tasks autonomously.

Installation

npm install -g @anthropic-ai/claude-code

Requires Node.js 18+ and an Anthropic API key (or Claude Pro/Max subscription).

Basic usage

Run it from inside your project directory:

claude

Claude Code automatically reads your project structure and can access any file in the working directory.

Key capabilities

Agentic task execution

Unlike chat tools, Claude Code can take actions: edit files, run tests, install packages, and iterate until a task is complete.

Fix the failing tests in src/auth/ and make sure all other tests still pass.

Full codebase context

Claude Code indexes your repository so you don't need to paste code manually. You can reference files by name:

Refactor the UserService class to use dependency injection.

Custom commands (slash commands)

You can define custom /commands in .claude/commands/ as Markdown files. These become reusable prompts available in any session. See the Skills page for details.

Hooks

Hooks are deterministic shell scripts that run outside the agentic loop in response to specific lifecycle events. Unlike prompts and instructions, hooks execute reliably every time the triggering event occurs.

Common hook events:

Event	When it fires	Typical use
PreToolUse	Before a tool is called	Block disallowed operations, validate inputs
PostToolUse	After a tool completes	Auto-format edited files, run linting
SessionStart	When a new session begins	Set up environment, display reminders
TaskCompleted	When a task finishes	Run tests, send notifications

Example: auto-format on save

Configure in .claude/settings.json:

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Edit|Write",
        "command": "npx prettier --write $CLAUDE_FILE_PATH"
      }
    ]
  }
}

Hooks are useful for guardrails that must always run (formatting, linting, security checks) rather than relying on instructions that the model might skip.

MCP servers

Claude Code supports the Model Context Protocol (MCP), allowing you to connect it to external tools: databases, APIs, internal systems. See the MCP Servers page for configuration details.

Learn more

• CLAUDE.md: give Claude Code persistent project context
• Managing Context: compaction, checkpointing, and context usage tips
• Permissions: control which tools can run and which files can be accessed
• CLI Flags: full reference for all command-line flags
• Debugging: troubleshooting and environment tips
• Tips: best practices for effective use