Overview
Claude Code is Anthropic’s official command-line interface1 that brings Claude directly into the terminal as an agentic coding assistant. Unlike the web-based Claude chat, Claude Code operates within your local development environment — it can read and edit files, run shell commands, search codebases, interact with git, and manage complex multi-step engineering tasks autonomously.
Key Findings
- Claude Code uses an agentic loop architecture2: the model receives context, decides which tools to invoke, observes results, and iterates until the task is complete.
- A sophisticated permissions model gives users fine-grained control over what actions Claude can take automatically vs. requiring approval.
- CLAUDE.md files provide persistent, project-scoped memory and instructions that survive across sessions.
- MCP (Model Context Protocol)3 integration allows Claude Code to connect to arbitrary external tools and data sources.
- Hooks enable users to inject custom shell commands at specific points in the agentic loop (before/after tool use, on session start, etc.).
- Supports headless/CI mode4 for non-interactive use in pipelines and automation.
Footnotes
References
- Claude Code Documentation
- Claude Code GitHub
- Model Context Protocol
- Anthropic API Documentation
- Claude Code SDK (npm)
Contents
| File | Description |
|---|---|
| architecture | Agentic loop, tool system, and context management |
| features | Core features: slash commands, memory, permissions |
| configuration | Settings, CLAUDE.md, hooks, and MCP servers |
| workflows | Common usage patterns and best practices |
| sdk-and-automation | Claude Code SDK, headless mode, and CI integration |