Claude Code

An agentic AI coding assistant — built for developers who want to delegate real coding tasks directly from their terminal.

Claude Code is a powerful command-line tool that brings Claude's intelligence to your development workflow. It can read and edit files, run shell commands, browse the web, manage tasks, and orchestrate multi-agent pipelines — all from within your terminal.

---

✨ Features

• Agentic Coding — Claude autonomously reads, writes, and edits code across your entire project
• Bash Execution — Run shell commands with safety checks and permission controls
• File Operations — Read, write, edit, and glob files with full context awareness
• Web Search & Fetch — Search the web and retrieve content to inform decisions
• Multi-Agent Orchestration — Spawn and coordinate sub-agents (agent swarms) for complex tasks
• MCP (Model Context Protocol) — Extend Claude with external tools and data sources via MCP servers
• Plugins & Skills — Install and run community or custom plugins and reusable skill bundles
• Notebook Support — Edit Jupyter notebooks directly
• Voice Input — Interact via voice with built-in STT (speech-to-text) support
• IDE Integrations — Works seamlessly with VS Code, JetBrains, and other editors
• Scheduled Tasks / Cron — Set up recurring or one-time automated agent runs
• Plan Mode — Review and approve Claude's action plan before execution
• Git & GitHub Integration — Commit, push, open PRs, and review code via CLI
• Cost & Token Tracking — Monitor usage and costs per session
• Memory & Context — Persistent session memory via CLAUDE.md files
• OAuth & Keychain — Secure authentication with keychain-backed credential storage
• Remote Sessions — Connect to and manage remote agent sessions
• Policy & Rate Limits — Configurable limits and managed settings for teams and enterprises
• Hooks System — Configure event hooks that fire before/after tool calls
• Theming — Customizable terminal themes and output styles
• LSP Integration — Language Server Protocol support for richer code awareness
• Worktree Support — Git worktree isolation for safe parallel experimentation
• Teleport — Jump between remote and local environments seamlessly
• Desktop Handoff — Hand off sessions between terminal and desktop app

---

🗂️ Project Structure

claude-code/
├── main.tsx                  # Entry point — CLI bootstrapping & command registration
├── Tool.ts                   # Base tool interface and builder
├── tools.ts                  # Tool registry and assembly
├── context.ts                # System context and CLAUDE.md loading
├── commands.ts               # CLI command definitions
├── tasks.ts                  # Task management helpers
├── history.ts                # Session history tracking
├── cost-tracker.ts           # Token and cost tracking
│
├── tools/                    # All built-in tools
│   ├── AgentTool/            # Spawn sub-agents
│   ├── BashTool/             # Execute shell commands (with AST security parsing)
│   ├── FileReadTool/         # Read files
│   ├── FileEditTool/         # Diff-based file editing
│   ├── FileWriteTool/        # Write new files
│   ├── GlobTool/             # File pattern matching
│   ├── GrepTool/             # Content search (ripgrep)
│   ├── WebFetchTool/         # Fetch and parse web content
│   ├── WebSearchTool/        # Web search
│   ├── MCPTool/              # MCP server tool calls
│   ├── NotebookEditTool/     # Jupyter notebook editing
│   ├── SkillTool/            # Run installed skills
│   ├── TodoWriteTool/        # Task list management
│   ├── ScheduleCronTool/     # Cron/scheduled tasks
│   ├── EnterPlanModeTool/    # Switch to plan mode
│   ├── EnterWorktreeTool/    # Git worktree isolation
│   ├── SendMessageTool/      # Send messages between agents
│   ├── TaskCreateTool/       # Create background tasks
│   ├── TaskListTool/         # List running tasks
│   ├── TaskStopTool/         # Stop running tasks
│   ├── TeamCreateTool/       # Create agent teams
│   ├── AskUserQuestionTool/  # Ask user clarifying questions
│   ├── LSPTool/              # Language Server Protocol queries
│   ├── REPLTool/             # Interactive REPL execution
│   ├── PowerShellTool/       # Windows PowerShell support
│   ├── SleepTool/            # Pause agent execution
│   ├── ToolSearchTool/       # Search available tools
│   └── SyntheticOutputTool/  # Synthetic output generation
│
├── commands/                 # All slash commands
│   ├── commit.ts             # /commit — Git commit with auto message
│   ├── review.ts             # /review — Code review
│   ├── model/                # /model — Switch AI model
│   ├── mcp/                  # /mcp — Manage MCP servers
│   ├── plugin/               # /plugin — Plugin management
│   ├── skills/               # /skills — Skills management
│   ├── hooks/                # /hooks — View & configure hooks
│   ├── theme/                # /theme — Change terminal theme
│   ├── voice/                # /voice — Toggle voice mode
│   ├── plan/                 # /plan — Enter plan mode
│   ├── memory/               # /memory — Manage CLAUDE.md memory
│   ├── session/              # /session — Session management
│   ├── tasks/                # /tasks — Background task management
│   ├── agents/               # /agents — View running agents
│   ├── cost/                 # /cost — Show cost breakdown
│   ├── diff/                 # /diff — Show file diffs
│   ├── compact/              # /compact — Compact conversation history
│   ├── export/               # /export — Export session
│   ├── share/                # /share — Share session
│   ├── doctor/               # /doctor — Diagnose issues
│   ├── login/logout/         # /login, /logout — Auth commands
│   ├── permissions/          # /permissions — View permission rules
│   ├── keybindings/          # /keybindings — Configure shortcuts
│   ├── config/               # /config — Edit settings
│   ├── ide/                  # /ide — IDE connection management
│   ├── upgrade/              # /upgrade — Upgrade Claude Code
│   ├── issue/                # /issue — Report issues
│   ├── feedback/             # /feedback — Send feedback
│   ├── stats/                # /stats — Session statistics
│   ├── status/               # /status — Connection status
│   └── vim/                  # /vim — Toggle vim mode
│
├── services/                 # Core services
│   ├── api/                  # Anthropic API client & bootstrap
│   ├── analytics/            # Telemetry, GrowthBook feature flags
│   ├── mcp/                  # MCP server connection management
│   ├── lsp/                  # Language Server Protocol
│   ├── oauth/                # OAuth authentication flows
│   ├── plugins/              # Plugin loading & registry
│   ├── policyLimits/         # Usage policy enforcement
│   ├── AgentSummary/         # Agent run summarization
│   ├── compact/              # Conversation compaction
│   ├── extractMemories/      # Memory extraction from sessions
│   ├── remoteManagedSettings/# Enterprise remote settings
│   ├── settingsSync/         # Settings synchronization
│   ├── tips/                 # In-app tip delivery
│   └── voice/                # Voice mode & STT
│
├── utils/                    # Utility modules
│   ├── auth.ts               # Authentication helpers
│   ├── config.ts             # Global config management
│   ├── git.ts                # Git utilities
│   ├── permissions/          # Full permission system
│   │   ├── PermissionMode.ts # Permission modes (default, plan, auto, etc.)
│   │   ├── bashClassifier.ts # Bash command safety classification
│   │   ├── dangerousPatterns.ts # Dangerous command detection
│   │   └── yoloClassifier.ts # Auto-accept classifier
│   ├── bash/                 # Bash AST parsing & security
│   ├── secureStorage/        # Keychain credential storage
│   ├── model/                # Model selection & routing
│   ├── startupProfiler.ts    # Startup performance profiling
│   └── ...
│
├── entrypoints/              # Application entry points
│   ├── cli.tsx               # Interactive REPL launcher
│   ├── init.ts               # Initialization logic
│   ├── mcp.ts                # MCP server entrypoint
│   └── sdk/                  # Agent SDK entrypoints
│
├── tasks/                    # Task execution layer
│   ├── LocalAgentTask/       # Local agent execution
│   ├── RemoteAgentTask/      # Remote agent execution
│   ├── LocalShellTask/       # Shell task management
│   ├── InProcessTeammateTask/# In-process teammate agents
│   └── DreamTask/            # Background dream tasks
│
├── screens/                  # Full-screen UI views
│   ├── App.tsx               # Root application screen
│   ├── Doctor.tsx            # Diagnostics screen
│   └── REPL.tsx              # Main REPL interface
│
├── components/               # Reusable UI components (Ink/React)
│   ├── Message.tsx           # Message rendering
│   ├── PromptInput/          # User input component
│   ├── StructuredDiff/       # Visual diff display
│   ├── Settings/             # Settings UI
│   ├── HighlightedCode/      # Syntax-highlighted code
│   ├── Markdown.tsx          # Markdown renderer
│   ├── StatusLine.tsx        # Bottom status bar
│   ├── ModelPicker.tsx       # Model selection UI
│   ├── ThemePicker.tsx       # Theme selection UI
│   ├── mcp/                  # MCP management UI
│   ├── permissions/          # Permission dialog UI
│   ├── tasks/                # Task list UI
│   └── ...
│
├── types/                    # TypeScript type definitions
├── schemas/                  # Zod validation schemas
├── bootstrap/                # App bootstrap & global state
├── coordinator/              # Multi-agent coordinator mode
├── plugins/                  # Plugin infrastructure
├── voice/                    # Voice mode implementation
├── vim/                      # Vim keybinding mode
└── memdir/                   # Memory directory management

---

🚀 Getting Started

Prerequisites

• Bun runtime (recommended) or Node.js 18+
• An Anthropic API key or Claude.ai subscription

Installation

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

# Or via Bun
bun install -g @anthropic-ai/claude-code

Authentication

# Log in with your Anthropic account
claude login

Basic Usage

# Start an interactive session in your project
claude

# Run a one-off task
claude "refactor the auth module to use async/await"

# Ask a question about your codebase
claude "explain how the permission system works"

# Commit changes with an auto-generated message
claude commit

# Review code changes
claude review

---

💬 Slash Commands

Inside an interactive session, you can use slash commands to control Claude Code:

Command	Description
/model	Switch the AI model
/plan	Enter plan mode — review before execution
/mcp	Manage MCP servers
/plugin	Manage plugins
/skills	Manage skills
/hooks	View and configure tool event hooks
/theme	Change the terminal theme
/voice	Toggle voice input mode
/memory	Edit CLAUDE.md memory files
/commit	Git commit with auto-generated message
/diff	Show pending file diffs
/review	Code review mode
/compact	Compact conversation history
/cost	Show session cost breakdown
/export	Export the session
/share	Share the session
/tasks	View background tasks
/agents	View running sub-agents
/session	Manage sessions
/permissions	View active permission rules
/keybindings	Configure keyboard shortcuts
/config	Edit Claude Code settings
/ide	Manage IDE connections
/doctor	Run diagnostics
/stats	Show session statistics
/status	Show connection status
/vim	Toggle vim keybinding mode
/login / /logout	Authentication
/upgrade	Upgrade Claude Code
/feedback	Send feedback
/issue	Report a bug
/help	Show help

---

🔐 Permission Modes

Claude Code has a configurable permission system to control how much autonomy Claude has:

Mode	Description
default	Ask before executing potentially dangerous operations
plan	Show a full plan and require approval before taking any action
acceptEdits	Auto-accept file edits, ask for shell commands
bypassPermissions	Skip all permission prompts (use with caution)
dontAsk	Never ask for permission — fully autonomous mode

Switch modes using /plan in session, or configure via settings.

---

🔧 Key Concepts

Tools

Claude Code uses a rich set of built-in tools to interact with your environment. Each tool has defined input schemas (via Zod), permission requirements, and React/Ink UI rendering. Tools cover file I/O, shell execution, web access, MCP calls, agent spawning, and more.

MCP (Model Context Protocol)

MCP servers extend Claude's capabilities with external integrations — databases, APIs, project management tools, Slack, GitHub, and more. Configure MCP servers globally or per-project in your settings.

Skills

Skills are reusable prompt bundles that give Claude specialized instructions for specific tasks (e.g., creating Excel files, PowerPoint decks, Word docs, or PDFs). Install community skills or author your own in a skills/ directory.

Plugins

Plugins bundle MCPs, skills, and tools together into installable .plugin packages for teams and organizations — making it easy to share workflows.

CLAUDE.md

Place a CLAUDE.md file in your project root (or ~/.claude/CLAUDE.md globally) to give Claude persistent memory, project-specific instructions, coding conventions, and context about your codebase.

Plan Mode

Enable plan mode (/plan) to have Claude show you its intended actions before executing them — giving you full review and control before any changes are made.

Hooks

Hooks fire before or after tool executions. They let you run custom scripts, logging, validation, or notifications at specific points in Claude's workflow.

Agent Swarms

Claude Code supports multi-agent orchestration where a coordinator agent spawns specialized sub-agents to work on different parts of a task in parallel, dramatically speeding up complex workflows.

Worktrees

The EnterWorktreeTool creates an isolated Git worktree so Claude can safely experiment with changes without affecting your main branch.

---

⚙️ Configuration

Claude Code stores its configuration in ~/.claude/. Key settings include:

Setting	Description
model	The Claude model to use (e.g., claude-opus-4-5, claude-sonnet-4-5)
permissionMode	Controls auto-approval of tool calls
mcpServers	MCP server configurations
env	Environment variable overrides
theme	Terminal color theme
outputStyle	Response formatting style
vim	Enable vim keybindings
hooks	Tool event hook configurations

---

🛠️ Tech Stack

Layer	Technology
Runtime	Bun
Language	TypeScript
CLI Framework	Commander.js
Terminal UI	Ink (React for CLIs)
Validation	Zod
AI SDK	@anthropic-ai/sdk
Search	ripgrep
Feature Flags	GrowthBook
Telemetry	OpenTelemetry
Bundler	Bun bundler with feature flags
Linting	Biome

---

🖥️ IDE Integrations

Claude Code integrates directly with popular editors:

• VS Code — Auto-connects and syncs file context
• JetBrains IDEs — IntelliJ, WebStorm, PyCharm, etc.
• Cursor — Native support
• Zed — Supported via LSP bridge

Connect your IDE using /ide inside a Claude Code session.

---

🌐 MCP Servers

MCP (Model Context Protocol) servers extend what Claude can access and do. Some popular integrations:

• GitHub — Read/write issues, PRs, code
• Slack — Read/send messages
• Databases — Query SQL/NoSQL databases
• File systems — Access remote file systems
• APIs — Connect to any REST or GraphQL API

Manage servers with /mcp or edit ~/.claude/settings.json.

---

📖 Documentation

• Official Docs
• Anthropic API Reference
• MCP Protocol Spec
• Prompting Guide

---

🤝 Contributing

Contributions, bug reports, and feature requests are welcome! Please open an issue or pull request.

---

📄 License

This source code is the property of Anthropic. All rights reserved. This repository is for personal and educational reference only and is not intended for redistribution or commercial use.