@anthropic-ai/claude-code. It lets you interact with Claude directly from your terminal to perform real software engineering work: editing files, running commands, searching codebases, managing git workflows, and more.
What problems does Claude Code solve?
Using a web-based LLM chatbot for coding requires you to copy-paste code back and forth, manually apply changes, and switch context constantly. Claude Code eliminates that friction by running inside your terminal, with direct access to your filesystem, shell, and version control. Claude can read your actual files, make edits, run tests, commit code, and handle multi-step workflows — all without leaving your editor or the command line.Key capabilities
File editing
Read, write, and edit files across your project. Claude uses context-aware edits and can handle multiple files in a single task.
Bash commands
Execute shell commands, run tests, install packages, and interact with your development environment directly.
Code search
Search your codebase with glob patterns and regular expressions using built-in
GlobTool and GrepTool capabilities.Git workflows
Stage files, commit, create branches, inspect diffs, and manage pull requests without leaving the CLI.
Multi-agent tasks
Spawn parallel sub-agents (
AgentTool) to tackle complex, multi-step engineering tasks concurrently.MCP integration
Connect external tools and data sources via the Model Context Protocol. List resources, call tools, and authenticate with MCP servers from inside a session.
Hooks and automation
Attach lifecycle hooks to tool calls so you can enforce policies, log actions, or trigger external systems automatically.
Session management
Resume previous conversations, run tasks in the background, and manage multiple concurrent sessions.
How it differs from using an LLM chatbot
| Chatbot | Claude Code |
|---|---|
| You copy-paste code in and out | Claude reads and writes your files directly |
| You run commands and paste output back | Claude runs commands and sees the output itself |
| Each message loses prior context | Sessions preserve full conversation history |
| No access to your git history or diffs | Claude can inspect commits, branches, and diffs |
| Single-threaded Q&A | Multi-agent: Claude can spawn sub-agents for parallel work |
| No external tool integrations | MCP servers connect Claude to databases, APIs, and more |
How Claude Code is invoked
Claude Code exposes theclaude binary. You can use it in two ways:
Interactive mode — a full REPL where you and Claude collaborate turn by turn:
-p / --print) mode — run a single prompt and get output to stdout, suitable for scripting and CI pipelines:
Next steps
Quickstart
Install Claude Code and run your first session in minutes.
Installation
Full installation guide including authentication options and platform notes.