@anthropic-ai/claude-code npm package v2.1.88, providing an in-depth reference for the architecture, tools, commands, and internal systems.
Quick Start
Get Claude Code running in minutes with authentication and your first query.
Installation
Install options: npm global, run pre-built CLI, or build from source.
Core Concepts
Understand the agent loop, tool system, permission model, and context management.
Architecture
Deep dive into the query engine, multi-agent system, MCP integration, and persistence.
Explore the reference
Built-in Tools
40+ tools for file operations, search, execution, agents, and MCP protocol.
Slash Commands
80+ slash commands for session management, configuration, and workflow control.
Configuration
Settings, environment variables, hooks, and the CLAUDE.md memory system.
Analysis Reports
Research findings: telemetry, hidden features, remote control, and roadmap.
Key capabilities
1
Agentic loop
Claude Code runs a continuous loop: send a prompt to Claude, receive a response, execute any requested tools, append results, and loop back — until Claude stops requesting tool calls.
2
Permission system
Every tool call passes through a layered permission system: pre-tool hooks, always-allow/deny rules, and interactive prompts. Configure permissions via
settings.json or CLI flags.3
Multi-agent orchestration
Spawn sub-agents via
AgentTool in fork, worktree, or in-process mode. Coordinate teams of agents sharing a task board and message inbox using TeamCreateTool and SendMessageTool.4
MCP integration
Connect to any Model Context Protocol server (stdio, SSE, HTTP, WebSocket) to extend Claude Code with custom tools and resources using the standard MCP protocol.
This repository contains decompiled source code for technical research and educational purposes. All source code is copyright Anthropic. Commercial use is prohibited.