Claude CodeNew

What Is Claude Code?

Claude Code is Anthropic’s command-line coding tool, powered by the Claude model series. You launch it in your project directory and issue tasks in natural language. It will:

• Understand the codebase: Automatically retrieve relevant files and build context without requiring you to manually paste snippets one by one.
• Execute multi-step workflows: Break down tasks, modify multiple files, run builds and tests, and self-correct based on error messages.
• Take action: Perform git operations, run scripts, and invoke command-line tools, creating a closed loop of "edit—run—verify."

Five Surfaces, One Setup

It started as a CLI only. The current documentation lists five surfaces: the terminal CLI, a VS Code extension, a JetBrains plugin, a desktop app (macOS and Windows, with Ubuntu and Debian in beta), and the browser at claude.ai/code plus the mobile app. The desktop app bundles Claude Code, so there is no separate CLI install; the web surface suits long-running tasks you kick off and check back on later.

All surfaces connect to the same engine, so a repository's CLAUDE.md, settings, and MCP servers apply everywhere, and sessions move between them: /desktop hands a terminal session to the desktop app for visual diff review, and a cloud task started with claude --cloud can be pulled back into the terminal with claude --teleport (both require a claude.ai subscription).

Getting Started: From Install to Your First Change

1. Install. On macOS, Linux and WSL, use the official install script (native installs update automatically in the background):

curl -fsSL https://claude.ai/install.sh | bash

On Windows, run irm https://claude.ai/install.ps1 | iex in PowerShell; you can also install with Homebrew (brew install --cask claude-code) or WinGet, though those don't auto-update. Run claude --version afterward to confirm.

1. Log in. Run claude in your project directory; the first time, it walks you through signing in to your Claude subscription or Console account in the browser. To switch accounts later, type /login in a session.
2. Let it understand the project first. Start with questions like "what does this project do?" or "where is the entry point?" and make sure it has it right before it changes anything.
3. Start with small changes. Have it change one function or add one test, review the diff it shows, and accept only once you're satisfied.

A few everyday commands: claude -c continues the most recent conversation in the current directory, claude -r lets you pick a past session to resume, and claude -p "question" runs a one-off query and exits; press Shift+Tab in a session to switch permission modes.

Typical Use Cases

• Implementing new features: Describe the requirements, and have it make cross-file changes while adding tests.
• Fixing bugs: Paste error logs or reproduction steps, and let it locate, fix, and verify the issue.
• Refactoring and migration: Perform batch changes, upgrade dependencies, and unify code style.
• Understanding unfamiliar codebases: Have it explain the call chains and data flow of a specific module.

Key Features

• Agent-based workflow: Excels at complex tasks requiring multiple steps and cross-file modifications—the key difference from standard completion tools.
• Strong context handling: Claude models demonstrate outstanding comprehension of large codebases and complex logic compared to peers.
• Controllability: Critical operations present plans and confirmation points to prevent unchecked, chaotic changes.
• Extensible: MCP connects external data sources (Google Drive, Jira, Slack); Skills package repeatable workflows into commands like /review-pr; Hooks run your own shell commands before or after actions, such as auto-formatting after every edit; subagents split work in parallel.
• Fits into pipelines: GitHub Actions and GitLab CI for PR review and issue triage, @Claude in Slack to turn a bug report into a pull request, and Routines for scheduled jobs that run in the cloud.

Usage Recommendations

When making significant changes, have it propose a plan before executing. Always run tests and perform code reviews afterward—AI makes mistakes, especially when unfamiliar with project conventions. Using it within a git repository allows for easy rollbacks, providing clearer safety boundaries. Treat it as a "high-level assistant that can take action," not as unreviewed automation.

Permissions and Safety

It can read and write files and run commands, and its permission settings decide what it can reach. You can block sensitive files with deny rules, such as Read(./.env) in settings, but the official documentation notes that such rules match command text and aren't a complete security boundary — for restrictions that apply to every process, turn on the operating system-level sandbox. Before using it in an unfamiliar repository, enable the sandbox and restrict network access; see our permissions checklist for details.

How It Compares

• If you already pay for ChatGPT and prefer OpenAI's ecosystem: OpenAI Codex; see our detailed comparison.
• If you prefer controlling changes line by line in an editor: Cursor; see Claude Code vs Cursor.
• For an open-source terminal agent where you choose the model: OpenCode.
• If the work isn't code but documents and spreadsheets: Claude Cowork.

Pricing

Claude Code is not part of the free tier; it requires a subscription or API access. The official pricing page (checked 2026-09-06) lists Pro at $20/month, or about $17/month billed annually; Max from $100/month in 5x and 20x tiers; Team standard seats at $25/month ($20 annually) and premium seats at $125/month ($100 annually); and Enterprise at a per-seat price plus API usage. All of those paid tiers include Claude Code. You can also use the Anthropic API directly and pay per token. Check the official page for current figures.