Getting Started

Claude Code is the official command-line tool from Anthropic that lets you interact directly with Claude in your terminal and have Claude help you complete various programming tasks.

What is Claude Code?

Claude Code is a powerful AI coding assistant that can:

• 📝 Read and write code - Directly read and modify your project files
• 🔍 Understand projects - Analyze codebase structure and dependencies
• 🛠️ Execute commands - Run tests, builds, deployments and other tasks
• 🤖 Work autonomously - Automatically complete complex tasks through Agent mode
• 🔗 Tool integration - Connect to various external tools via the MCP protocol

Quick Installation

macOS / Linux

# Install using Homebrew
brew install anthropics/claude/claude

# Or install using npm
npm install -g @anthropic-ai/claude-code

Windows

# Install using npm
npm install -g @anthropic-ai/claude-code

First Use

1. Get an API Key

Visit Anthropic Console to create an API Key.

2. Configure API Key

# Set environment variable
export ANTHROPIC_API_KEY='your-api-key-here'

# Or set in configuration
claude config set apiKey your-api-key-here

3. Launch Claude Code

# Launch in your project directory
cd your-project
claude

# Or specify a project path
claude /path/to/project

First Conversation

After launching, you'll see the Claude Code interactive interface:

Claude Code v2.1.0
Working directory: /Users/you/project

You: Help me analyze this project's structure

Claude will automatically:

1. Read project files
2. Analyze the code structure
3. Provide a detailed project overview

Basic Commands

Built-in Commands

/help          # View help
/clear         # Clear conversation history
/quit          # Exit Claude Code
/model         # Switch model (sonnet/opus/haiku)
/cost          # View API usage cost
/fast          # Toggle fast mode (same model but faster output)

Keyboard Shortcuts

Shift+Tab - Cycle through permission modes:

Normal Mode → Auto-Accept Mode → Plan Mode → Normal Mode

• Normal Mode: Default mode, each operation requires confirmation
• Auto-Accept Mode: Automatically accept edits (bottom shows ⏵⏵ accept edits on)
• Plan Mode: Read-only mode (bottom shows ⏸ plan mode on), Claude only plans without executing

Common Operations

# Have Claude read a file
You: Read the src/index.ts file

# Have Claude modify code
You: Add a login button to App.tsx

# Have Claude run tests
You: Run all unit tests

# Have Claude explain code
You: Explain what utils/helper.js does

Working Modes

1. Normal Mode (Default)

Chat directly with Claude, each operation requires confirmation:

You: I want to add user authentication
Claude: I'll help you implement that. Let me first look at the existing code structure...

Claude wants to:
  • Read file: src/auth.ts
  • Write file: src/auth/login.ts
Allow? [y/n/always/never]

2. Auto-Accept Mode

Press Shift+Tab to switch to auto-accept mode, Claude will automatically execute all edit operations:

Bottom shows: ⏵⏵ accept edits on

You: Add login functionality
Claude: I'll implement that...
[Automatically reads and writes files without confirmation]

Use cases: Trusted tasks, rapid iteration

3. Plan Mode

Press Shift+Tab to switch to plan mode, Claude only plans without executing:

Bottom shows: ⏸ plan mode on

You: Refactor the database layer
Claude: Let me create a refactoring plan...

Plan:
1. Create a new database abstraction layer
2. Migrate existing queries
3. Update tests
4. Verify performance

[No operations executed, only displays the plan]

Use cases: Complex tasks, when you need to review the plan first

Exit Plan Mode: Press Shift+Tab again to return to Normal Mode

Permission System

Claude Code has three permission modes that can be cycled through with the Shift+Tab shortcut:

Normal Mode (Default)

Asks for confirmation before each operation:

Claude wants to:
  • Read file: src/config.ts
  • Write file: src/auth.ts
  • Run command: npm test

Allow? [y/n/always/never]

Response options:

• y - Allow this operation
• n - Deny this operation
• always - Always allow this type of operation for this session
• never - Always deny this type of operation for this session

Auto-Accept Mode

Press Shift+Tab to switch, automatically accepts all edit operations:

Bottom shows: ⏵⏵ accept edits on

Features:

• Automatically executes file reads and writes
• Still asks about dangerous operations (e.g., deleting files, git push)
• Suitable for rapid iteration and trusted tasks

Plan Mode

Press Shift+Tab to switch, read-only mode:

Bottom shows: ⏸ plan mode on

Features:

• Claude only creates plans, does not execute operations
• Safely explore complex tasks
• After confirming the plan, exit Plan Mode to execute

Command-line Startup Options

# Start in Auto-Accept mode
claude --permission-mode auto

# Start with custom permission configuration
claude --permission-mode custom

Custom Permission Configuration

Configure rules in .claude/settings.json:

{
  "permissionMode": "custom",
  "allowedPaths": ["src/**", "tests/**"],
  "allowedCommands": ["npm test", "npm run build"],
  "blockedPaths": [".env", "secrets/**"]
}

Project Configuration

Create a CLAUDE.md file in the project root to tell Claude important project information:

# Project Description

This is a React + TypeScript project.

## Tech Stack
- React 18
- TypeScript 5
- Vite
- TailwindCSS

## Code Standards
- Use functional components and Hooks
- All components must have TypeScript types
- Use ESLint and Prettier

## Testing
Run tests: `npm test`

Next Steps

• Installation - Detailed installation and configuration guide
• First Conversation - Learn how to communicate effectively with Claude
• Getting Started - Master common features and tips
• Workflows - Learn efficient workflows

FAQ

Claude can't read my files?

Check your permission settings and file paths. Claude can only access the current working directory and its subdirectories.

How do I make Claude remember project information?

Create a CLAUDE.md file with project descriptions, tech stack, code standards, and other information.

How are API costs calculated?

Use the /cost command to view the current session's costs. Claude Code uses the Anthropic API, billed by token.

Can I use it in CI/CD?

Yes! Use Headless mode:

claude --headless "Run tests and generate report"

Getting Help

• 📖 Full Documentation
• 💬 GitHub Discussions
• 🐛 Report Issues
• 🔧 Command-line help: claude --help