What is Claude Code?

What Claude Code is

Claude Code is a coding agent from Anthropic that runs in your terminal. It reads the files in a real repository, edits them, runs commands such as npm test or git commit, and reads the output to decide what to do next. You describe a task in plain language and it does the work in your working directory.

A chat window can only hand you text to copy back into your editor. Claude Code holds the file system and the shell, so it changes the code, runs the tests, sees the failure and fixes it without you carrying text between two windows.

Starting a session on macOS, Linux or WSL (Windows Subsystem for Linux) takes three commands:

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

The first run opens your browser to log in. After that, claude --version prints a version number followed by (Claude Code). If your shell answers claude: command not found instead, the launcher was installed to ~/.local/bin, and a shell you opened before the install does not have that directory on its PATH yet. Open a new terminal and run it again. claude doctor prints installation and settings diagnostics without starting a session, which is the fastest way to see what a broken install is missing.

Where does Claude Code run?

The terminal is the main surface. It supports macOS 13.0 and newer, Windows 10 1809 and newer, Ubuntu 20.04 and newer, Debian 10 and newer, and Alpine 3.19 and newer, on x64 or ARM64 hardware with at least 4 GB of RAM. Bash, Zsh, PowerShell and CMD all work.

The same engine drives the other surfaces: extensions for VS Code and JetBrains IDEs, a desktop app, the browser at claude.ai/code, and the Claude mobile app. Your repository's CLAUDE.md file, your settings and your MCP (model context protocol) servers apply on all of them, so moving between surfaces does not mean configuring anything twice.

An internet connection is required, because the model itself runs on Anthropic's servers. Your machine runs only the agent loop: the file reads, the edits and the shell commands. Nothing here works offline.

That split is why a small server makes a good host. A long task holds a terminal open for as long as it runs, so people put the agent on a VPS, start it inside tmux over SSH, and then close the laptop without killing the session. Running Claude Code on a VPS inside tmux walks through that setup. A box with room for one session usually has room for several, and one session can pass text straight to another rather than waiting for you to reattach and carry the result across yourself. For a machine sitting in front of you instead, Claude on the Linux desktop and the CLI covers the local install.

What does Claude Code cost?

There are two ways to pay for it, and they behave differently.

A Claude subscription includes Claude Code. The Pro, Max, Team and Enterprise plans all carry it, but they do not bill it the same way. On Pro, Max and Team your usage is bounded by the plan's rolling limits rather than by a bill that keeps growing. When you reach a limit you can wait for the window to reset. On Pro and Max you can also turn on usage credits, an opt-in that keeps requests going at standard API rates instead of blocking you. They are charged on top of the subscription from a prepaid balance, with a monthly cap you can set, and they cover Claude Code in the terminal as well as chat. Anthropic's pricing page lists the same extra usage for Team. Enterprise is the exception: as of 13 October 2026 its seat fee covers access only, and every token is billed at API rates. Max sells at two prices for the same models and the same features. The tiers differ in how much usage each window allows, and in the monthly Console API credits they include ($100 on Max 5x, $200 on Max 20x), which Anthropic says cannot be spent in Claude Code or the Claude apps. So choosing between the $100 and $200 tiers comes down to how often you are the one waiting, or paying for usage credits.

An Anthropic Console account bills per token instead. There is no usage window to wait out, only the monthly spend cap of your Console tier or a lower limit you set, so a long agent session on a large repository costs real money in real time. That is the honest cost driver here: the agent pays for every token it reads, and an agent exploring a big codebase reads a lot of it. The same task in a small, tidy repository costs less than in a sprawling one.

Buying for a group is a third case, because Team and Enterprise are sold by the seat, and the point where per-token billing undercuts a row of seats depends on how many of those people run long sessions rather than how many names are on the account.

Plan prices move, so beyond the two Max prices above, checked on 13 October 2026, this page quotes none of them. What Claude Code costs per month carries the current numbers and the arithmetic for choosing between a subscription and per-token billing. Knowing the price is only half the question, since whether a plan pays for itself depends on your own hourly rate and how many hours the agent actually saves you.

Is Claude Code free?

No. As of October 2026, Claude Code requires a Pro, Max, Team, Enterprise or Console account, and the free Claude.ai plan does not include it. Signing in with a free account gives you the chat product without the agent.

There is no free tier of the agent and no offline mode, because every turn is a request to a model running on Anthropic's servers. The one free route is a guest pass: eligible Pro and Max subscribers have a limited number to share, and each gives someone new to paid Claude seven days of the Pro plan, Claude Code included. Without one, the cheapest honest way to try it is one month of the entry paid plan. Whether Claude Code is included with Claude Pro goes through what a Pro plan does and does not cover.

What Claude Code is not

It is not a chat window. It edits files and runs commands on your machine, which is why it asks permission before actions that change something, and why many people run it in a container or on a server they can rebuild. It does not talk back at length either, since the transcript is closer to a work log than a conversation, and an output style rewrites the system prompt behind it when you want the agent to explain its reasoning as it goes.

It is not an IDE. There is no editor pane and no file tree. It attaches to the editor you already use, or to no editor at all.

It is not an autocomplete tool. It does not offer the next line while you type. The unit of work is a whole task, such as "find why this test is flaky and fix it". Handing over a whole task also hands over the scope of it, so when the agent keeps rewriting three files where one line would do, a rule like the Ponytail skill pushes it back toward the smallest change that works.

It is not a local model. The weights are not on your disk, so a machine with no network connection runs nothing.

Readers often ask for a ChatGPT comparison at this point. That is the wrong axis. The chat products answer questions; Claude Code changes a repository. The comparison worth reading is against the other agents that also hold your file system, and Claude Code against Cursor, Codex and Copilot is that comparison.

FAQ

Is Claude Code free to use?

No. As of October 2026 it needs a paid Claude plan (Pro, Max, Team or Enterprise) or an Anthropic Console account billed per token. The free Claude.ai plan gives you the chat product without the agent. The only free trial is a guest pass from an eligible Pro or Max subscriber, which gives someone new to paid Claude seven days of Pro with Claude Code included. There is no offline mode, since every turn is a request to a model running on Anthropic's servers.

Do I need to know the terminal to use Claude Code?

A little. You need to open a terminal, change directory into your project, and run claude. From there you type in plain language. If the terminal is the blocker, the desktop app and the browser version at claude.ai/code give you the same agent behind a graphical interface, but neither accepts every account the terminal does. Both need a Claude subscription rather than a Console API key. The desktop app's Code tab asks for Pro, Max, Team or Enterprise, and the browser version is open to Pro, Max and Team plans and to Enterprise users with premium or Chat + Claude Code seats, signed in with a claude.ai account.

Does Claude Code send my code to Anthropic?

The parts it reads, yes. The model runs on Anthropic's servers, so any file the agent opens is sent as part of that request. Files it never opens are never sent. If that matters for a repository, start the agent in a directory that holds only what you are willing to send.

Can I leave Claude Code running while my computer is off?

Not on your own machine, because a terminal session dies with the terminal that owns it. Two ways around that. Run it on a VPS inside tmux, where the session survives the SSH disconnect and you reattach later. Or start the task from the browser or the mobile app, where it runs on Anthropic's infrastructure instead of yours. That route needs a claude.ai sign-in on a plan that includes cloud sessions, not a Console API key.