What Is Claude Code?
An agentic coding assistant that lives in your shell, reads your whole project, and runs commands with your approval. Here is how it works.
The copy-paste loop is brutal. You grab a snippet, drop it into a browser AI, wait, paste the fix back into your editor, watch it break something else, and start over. That loop is exactly what Claude Code kills.
Quick Win: See It Work in 2 Minutes
Point your terminal at any project folder. Then run:
npx @anthropic-ai/claude-code
claude "explain this codebase structure and find potential issues"No copy-pasting. No context loss. Your files, your Git history, and your dependencies all get pulled in, and Claude hands back actionable notes. It already knows the project.
What Is Claude Code?
Claude Code is Anthropic's agentic coding assistant. It runs in your terminal, your IDE, or your browser. Unlike chat AI that lives behind a textbox, it has real access to your project, runs commands on your machine, and makes changes once you approve them.
Here is the shift: you are now the main thread. Shuttling context by hand is over. You queue up tasks and Claude chews through them. Picture yourself as a CPU scheduler, handing work to a worker that runs jobs in parallel.
# Claude Code works where you work
claude "add authentication to the Express routes"
claude "write tests for the payment service"
claude "refactor this component to use React hooks"How It Works
Everything runs locally. Claude talks straight to Anthropic's API. Your code is never indexed remotely and never processed on a server. Three capabilities carry the weight:
Full Codebase Awareness: File tree, Git history, and dependencies get picked up automatically. Ask about any file and the context is already loaded.
Agentic Execution: Code suggestions are only half of it. Claude writes files, runs commands, makes commits, and verifies its own work. Every action waits on your approval.
Extensibility: MCP servers plug Claude into databases, APIs, GitHub, and other outside tools so it can reach beyond your repo.
Key Features You Should Know
Plan Mode (shift+tab twice): A read-only analysis pass. Useful when you want to understand a thorny problem before anything gets touched.
Skills: Packaged instructions Claude loads on demand. They teach it a workflow for one domain.
Hooks: Event-triggered actions. Format code after every edit, or block writes to a production file.
Checkpoints: Automatic snapshots before each change. Esc twice rewinds when something breaks.
Subagents: Parallel workers for heavy tasks. Spin up one on the frontend, another on the backend, and let them run at the same time.
Claude.ai vs Claude Code
| Browser Chat | Claude Code |
|---|---|
| Copy-paste context manually | Reads your codebase automatically |
| Forgets between sessions | Maintains project context |
| Suggests code snippets | Executes changes with approval |
| Isolated from your tools | Integrates Git, npm, test runners |
The gap is not small. One is an advisor. The other is an engineer.
Getting Started
- Install: Grab the native installer for a one-command setup, or walk through the full install guide.
- First Project: Run the first-project tutorial once, end to end.
- Troubleshooting: Hit a snag? The common-fixes page covers most of them.
- Go Deeper: Read up on permission management when you want safer automation.
Pro and Max subscriptions both include Claude Code. Pay-as-you-go through the API works too. Opus 4.5, Sonnet 4.5, and Haiku 4.5 are all supported.
Grab the native installer for a clean setup, or check the full install guide if you need options.
Stop configuring. Start building.