What Is Claude Code? How to Install It Safely and Use Anthropic’s AI Coding Agent

Claude Code is Anthropic’s command-line AI coding agent, a terminal tool that reads your codebase, plans changes, executes them with real development tools, and iterates on the results. Unlike IDE assistants that autocomplete one line at a time, Claude Code works at the project level: it understands your whole repository, reasons through multi-file changes, and can handle complex engineering tasks with limited supervision.

If you have used Claude the chatbot for coding help and wished it could directly edit your files, run tests, and commit code, that is what Claude Code does.

How is Claude Code different from Copilot and Cursor?

Claude Code differs from Copilot and Cursor structurally: it runs in your terminal as an autonomous agent, rather than as autocomplete in your editor. The AI coding market has split into three shapes, and they are IDE extensions, AI-native editors, and terminal agents.

The difference shows up in what each does with the same request. Ask GitHub Copilot to “add error handling to this function,” and it suggests code for the function on your screen. Ask Claude Code the same thing, and it scans the whole project for functions that lack error handling, plans which ones to change, edits them across multiple files, runs your test suite to check nothing broke, and shows you the results. For how it stacks up against Copilot, Cursor, and the rest, see our best AI coding assistants roundup and our ChatGPT vs Claude vs Gemini for coding comparison.

Claude Code runs on Anthropic’s Claude models. As of October 2026, Claude Code defaults to Claude Opus 5.5 on Pro, Max, Team, and Enterprise accounts and on the Anthropic API, according to Anthropic’s model configuration documentation. Anthropic released Opus 5.5 on September 22, 2026, and the new default replaced the earlier split between Claude Sonnet 5 and Claude Opus 5. Opus 5.5 runs with a 1M-token context window by default, which lets Claude Code reason across large codebases. An organization admin can set a different default model, and you can switch models within a session. Anthropic reports strong results for these models on coding benchmarks such as SWE-bench (self-reported); independent testing varies, and benchmark scores are a rough guide rather than a guarantee for your codebase.

Who is Claude Code for?

Claude Code is built for developers comfortable in the terminal, and it is not a beginner tool. It assumes you know git, package managers, build systems, and basic software architecture.

Newer programmers who already have the fundamentals can still get value from it once they are ready to review changes carefully, run tests, and follow why the agent makes specific edits. If that is you, start with our guide to learning to code with AI first. Claude Code is particularly strong for backend work in Python, Node.js, TypeScript, and Go, for full-stack refactoring, for writing test suites, for codebase migrations such as JavaScript to TypeScript, and for debugging issues that span multiple files.

How do you install Claude Code?

Install Claude Code with Anthropic’s native installer, which is now the recommended path for most users. The old npm method still works but is now the advanced option, and it carries a Node.js version requirement the native installer avoids.

Install it in four steps:

1. Run the native installer. On macOS, Linux, or WSL, run curl -fsSL https://claude.ai/install.sh | bash. On Windows, run irm https://claude.ai/install.ps1 | iex in PowerShell, the CMD version of the script from the same setup page, or winget install Anthropic.ClaudeCode; both produce a native Windows binary, so WSL is not required. Per Anthropic’s official Claude Code setup documentation, native installations update automatically in the background, while WinGet and Homebrew (brew install --cask claude-code) installations must be upgraded manually.
2. Authenticate. Run claude and it opens your browser to log in to your Anthropic account. Credentials are stored locally.
3. Open your project. Change into the root of any repository. Claude Code detects your project structure, reads your package.json or equivalent, and indexes relevant files.
4. Start a session. Type claude to launch the interactive session, then describe what you want in plain language.

If you prefer the npm route for CI pinning or team standardization, run npm install -g @anthropic-ai/claude-code, which now requires Node.js 22 or newer. Do not install it with sudo, and if you switch from npm to the native build, remove the npm copy with npm uninstall -g @anthropic-ai/claude-code so the two versions don’t conflict on your PATH. Run claude doctor if anything misbehaves; it checks your PATH, authentication, and install type.

Claude Code is also available outside the terminal. Anthropic’s Claude Code overview lists a VS Code extension (which also installs in Cursor), a JetBrains plugin, a Code tab in the Claude desktop app for macOS and Windows, and a web version at claude.ai/code that runs tasks on Anthropic-managed cloud machines. The desktop app includes Claude Code, so it does not need a separate command-line install, but it still requires a paid subscription. For most developers, the practical choice is the surface that matches where they already work, because the same project settings and CLAUDE.md files apply across all of them.

Where can you download Claude Code safely?

Download Claude Code only from Anthropic. The official sources are Anthropic’s setup documentation at code.claude.com, whose install scripts download from claude.ai; the WinGet and Homebrew packages named on that page; the npm package @anthropic-ai/claude-code; and the Claude desktop app from claude.com. Any other website offering a “Claude Code download” is not an official source, however closely it copies Anthropic’s design.

This matters because Claude Code is a popular target for malware. In March 2026, Malwarebytes reported on cloned Claude Code install pages, promoted through sponsored search results, that kept everything identical except the one-line install command, which instead downloaded an infostealer that grabs saved passwords, cookies and session tokens. Researchers at Straiker later counted more than 88 fake domains impersonating Claude Code and other developer tools. Lookalike addresses built around “claudecode”, such as sites on .us.org, .us.com or .de.com domains, are not listed by Anthropic as official sources, so do not paste install commands from them.

Before you run any install command, check three things:

1. Get to the docs yourself. Type code.claude.com into your browser or follow the link from claude.com, rather than clicking a sponsored search result.
2. Read the command. The official one-liners fetch from https://claude.ai/install.sh, install.ps1 or install.cmd. A command that downloads from any other domain is not Anthropic’s.
3. Confirm the install. Open a new terminal and run claude --version, then claude doctor to check the install type and authentication.

If you already ran an install command from an unofficial site, disconnect the computer from the network, then from a different, clean device change the passwords and revoke the sessions and tokens for your email, GitHub, cloud and API accounts. Run a full scan with a reputable security tool, and consider reinstalling the operating system if the scan finds anything, since infostealers are built to take credentials quickly.

What can you actually do with Claude Code?

Claude Code handles whole tasks rather than snippets, and the clearest way to see that is by example. These are the jobs it does well, and they are understanding a codebase, multi-file refactoring, writing tests, debugging, and git work.

• Understand a new codebase. Ask it to explain the architecture, the main modules, how they interact, and where the entry point is. It reads the files, analyzes imports and dependencies, and returns a clear summary.
• Refactor across files. Ask it to switch the authentication module from session cookies to JWT tokens and update every route, the middleware, and the tests. It plans the change, edits the files, fixes imports, and adjusts tests, asking permission before each file edit by default.
• Write tests. Ask for unit tests covering all public methods and edge cases using your existing Jest config. It reads your test setup and follows your project’s conventions.
• Debug. Tell it a specific endpoint returns a 500 error when a subscription has expired. It traces the request flow, finds the root cause, explains it, and proposes or implements the fix.
• Work with git. Ask it to review your branch and write a commit message, summarize the diff against main, or draft a pull request description from your commits.

What are Claude Code’s main features?

Claude Code’s defining features are its cautious default, project memory, parallel agents, and MCP support. Together they make it configurable to how your project actually works.

Cautious by default means it asks permission before modifying files or running commands, showing you the plan and waiting for confirmation. You can loosen this, but the default is conservative. Project memory lets it store context such as coding conventions and architecture decisions in a project configuration file, so it improves as it learns your repository. On higher-tier plans, it can run parallel agents that work on different parts of a task at once, such as one refactoring the API while another updates the tests. And its support for the Model Context Protocol (MCP) lets it connect to external tools such as databases, APIs, and monitoring systems, extending it beyond reading and writing code.

How much does Claude Code cost?

Claude Code requires a paid Anthropic plan and has no free tier. Pricing is measured in tokens processed, so complex tasks on large codebases cost more.

As of October 2026, US plan pricing on Anthropic’s pricing page runs from Claude Pro at $20/month ($17/month billed annually), through Claude Max at $100/month for 5x Pro usage and $200/month for 20x. Team plans cost $25 per Standard seat per month ($20 billed annually) or $125 per Premium seat ($100 billed annually), and Enterprise is $20 per seat per month billed annually with a 20-seat minimum, plus usage billed separately at API rates. Confirm current pricing on Anthropic’s own plans page before you buy, since it changes. For a typical few-hour session, Pro is usually enough; developers who run Claude Code as their main workflow benefit from a Max plan. You can also authenticate with an API key by setting ANTHROPIC_API_KEY, but that bills API usage separately and can cost more than a subscription for heavy use.

What are Claude Code’s limitations?

Claude Code’s main limitations are that it is not an as-you-type autocomplete tool, it is single-vendor, and it is slower on very large tasks. Each one has a straightforward workaround or trade-off.

Claude Code works through instructions and reviewed changes rather than inline suggestions: the command-line version has no autocomplete as you type, and Anthropic describes its VS Code extension in terms of inline diffs, @-mentions, plan review, and conversation history rather than completion, so pair it with a separate editor tool if you want autocomplete. Single-vendor means no bring-your-own-model: you are on Anthropic’s Claude models, and if you want model flexibility, open-source tools such as Cline or Aider let you use any model via API keys. And large tasks take time, because processing a big context window is not instant, so a lightweight tool like Copilot is faster for quick questions or simple autocomplete. None of these is a dealbreaker, but they define where Claude Code fits: deep, multi-file work rather than fast single-line help.

Is Claude Code safe to use on proprietary code?

Claude Code can be used on proprietary code, but whether your code may be used for model training depends on your account type. According to Anthropic’s Claude Code data-usage documentation, Anthropic does not train generative models on code or prompts sent under commercial terms (Team, Enterprise, the API, and third-party cloud platforms) unless the customer opts in. For consumer Free, Pro, and Max accounts, Anthropic trains new models on Claude Code data when the model-improvement setting is on and keeps that data for 5 years; with the setting off, retention is 30 days.

This distinction matters because many developers use a personal Pro or Max account for work. Check the setting at claude.ai/settings/data-privacy-controls before you open employer or client code, and follow your employer’s policy, which may require a Team or Enterprise account. Zero data retention is available to qualified Enterprise accounts and is enabled per organization by Anthropic’s account team.

The cautious default matters most here: because Claude Code asks before editing files or running commands, you see and approve each change before it touches your repository. Treat that as a feature to keep enabled on proprietary or security-sensitive code rather than something to switch off for speed. If you work on security-critical systems, our guides to developer-first security platforms and agentic AI security risks are worth reading alongside this, since an autonomous agent editing your code is exactly the kind of tool those risks apply to.

FAQ

Is Claude Code free?

No. Claude Code requires a paid Anthropic plan starting at $20/month for Claude Pro, and the free Claude.ai plan does not include it. You can use the free Claude chatbot for coding questions, but not for the agentic file-editing, test-running, and git capabilities that define Claude Code.

Which model does Claude Code use?

Claude Code defaults to Claude Opus 5.5 on Pro, Max, Team, and Enterprise accounts and on the Anthropic API as of October 2026, with a 1M-token context window. You can switch models within a session, and an organization admin can set a different default.

Can I use Claude Code with VS Code?

Yes, Claude Code works with VS Code through Anthropic’s official extension, which adds inline diffs, @-mentions, plan review, and conversation history inside the editor. You can also run the command-line version in VS Code’s integrated terminal, and Anthropic offers a JetBrains plugin as well.

How do I install Claude Code?

Use Anthropic’s native installer, now the recommended path. On macOS, Linux, or WSL, run the install script from claude.ai; on Windows, use the PowerShell installer or WinGet package, which no longer require WSL. The older npm install still works for CI and team setups but needs Node.js 22 or newer. Run claude doctor to check your installation.

Is claudecode.us.org the official Claude Code site?

No. Anthropic’s official Claude Code documentation is at code.claude.com, and its install scripts download from claude.ai. Lookalike “claudecode” sites on other domains, including .us.org and .us.com addresses, are not listed by Anthropic as official sources. Fake Claude Code install pages were used to spread password-stealing malware in 2026, so install only from Anthropic’s own pages.

Where can I download Claude Code safely?

Use Anthropic’s setup page at code.claude.com, which gives install commands that download from claude.ai, plus official WinGet, Homebrew and npm options. The Claude desktop app from claude.com also includes Claude Code. Avoid sponsored search results and any page whose install command downloads from a different domain.

How is Claude Code different from asking Claude in the chatbot?

The Claude chatbot can discuss code, explain concepts, and generate snippets. Claude Code reads your project files, makes changes, runs commands, executes tests, works with git, and edits multiple files in a coordinated way. It is the difference between asking an expert for advice and having them sit at your desk and do the work.