What Is Claude Code? A Plain Explanation

Claude Code is Anthropic’s agentic coding tool. You describe what you want in plain language, and Claude reads the files in your project, edits code, runs commands such as your tests or git, looks at the results and keeps going until the job is done or it needs you. It runs in a terminal, in VS Code and JetBrains IDEs, in the Claude desktop app and in the browser at claude.ai/code. It is not free: it needs a paid Claude plan (Pro, Max, Team or Enterprise) or a Claude Console account. The difference from chatting with Claude is that Claude Code acts on your files instead of handing you text to copy.

What Claude Code is, in one paragraph

Anthropic’s page on how Claude Code works↗ (opens in a new tab) describes an agentic loop in three phases: gather context, take action, verify results. A Claude model does the reasoning; Claude Code is the layer around it, sometimes called the harness, that gives the model tools and manages what it sees. The tools are what make it an agent rather than a chatbot: it can read and create files, search the codebase, run shell commands, fetch web pages and, with the right plugins, see type errors after an edit. Each result feeds the next step, so “fix the failing tests” becomes run the tests, read the failure, find the code, change it, run the tests again.

If the word agent is new to you, what an AI agent is covers the idea in general, and what agentic AI means covers the wider trend.

Where it runs

The same engine sits behind several front ends, so your CLAUDE.md files, settings and MCP servers carry across them. Anthropic’s platforms page↗ (opens in a new tab) lists these today:

• Terminal: the claude command-line tool, the full feature set, and the one you script and run in CI.
• VS Code: an extension with inline diffs and plan review. It also installs in Cursor. See Claude Code in VS Code.
• JetBrains: a plugin for IntelliJ IDEA, PyCharm, WebStorm and the rest, which uses a separately installed CLI. See Claude Code in JetBrains.
• Desktop: the Code tab in the Claude app for macOS and Windows, with a Linux beta on Ubuntu and Debian. It includes Claude Code, so there is nothing else to install.
• Web: claude.ai/code runs each task in a cloud machine against a GitHub repository, and keeps going after you close the laptop. See Claude Code on the web.
• Mobile: the Claude app for iOS and Android can start cloud sessions and steer a session running on your own computer through Remote Control.
• Team tools: @Claude in Slack, and GitHub Actions or GitLab CI/CD for reviews and triage in a pipeline.

Which front end to pick is mostly taste; Desktop app vs CLI vs VS Code compares the three most people choose between.

What it can do

• Explain a codebase: where the entry point is, how a request flows, why a test is flaky.
• Build features and fix bugs across several files, then run the tests or the build to check its own work.
• Handle the chores: tests for untested code, lint fixes, dependency updates, merge conflicts, release notes.
• Work with git: stage changes, write commit messages, create branches and open pull requests.
• Use your other tools through MCP, such as an issue tracker, a design file or a task board.
• Follow your conventions: it reads a CLAUDE.md file at the start of every session, and you can add skills, hooks and subagents on top.
• Run without a person at the keyboard: claude -p takes a prompt and prints the result, so it fits into scripts and CI, and scheduled runs can happen in the cloud or on your machine.

# Interactive: start in your project folder
cd my-app
claude

# One-shot, with a task in the command
claude "write tests for the auth module, run them, and fix any failures"

# Scripted: pipe input in, print the answer
git diff main --name-only | claude -p "review these changed files for security issues"

What it cannot do, or will not do for you

• Remember yesterday on its own. Each new session starts with an empty context window. What should persist goes in CLAUDE.md or its auto memory; Claude Code memory explains both.
• Undo everything. Checkpoints restore files Claude edited, but not a database it changed, an API it called or a deployment it ran. Git is still your real undo.
• Guarantee the code is right. It is usually good and sometimes confidently wrong. Tests it can run, and your own read of the diff, are what catch the difference.
• Work offline. It needs an internet connection and runs only in the countries Anthropic supports.
• Keep a shared record of the work. Its to-do list lives inside one session, and the transcript is on one machine.

How much it does without asking depends on the permission mode. Recent versions start interactive sessions in auto mode, where a classifier reviews most actions in the background and blocks risky ones; Manual mode asks before every edit and command, and plan mode only reads and proposes. Auto-approve in Claude Code walks through each mode.

How it differs from chatting with Claude

In a chat at claude.ai you paste code in, Claude writes an answer, and you copy it back and try it. You are the one who runs things. Claude Code closes that loop: it opens the files itself, makes the edit where it belongs, runs the command, reads the error and tries again. The model can be the same one. What changes is that it has hands, and a working folder to use them in.

Two other Anthropic products get confused with it. Claude Projects in the chat app group conversations and files for reference, as Claude Projects vs Claude Code explains. Cowork, in the desktop app, runs longer jobs across your files and apps for people who do not code; Cowork vs Claude Code draws the line.

Who it is for

• Developers, first of all: people who can read a diff and judge whether a change is right.
• Teams that want shared settings. Team and Enterprise plans add admin controls and settings managed centrally; Claude Code for teams covers the setup.
• Product managers and analysts who work near code, as Claude Code for product managers describes.
• People who do not code at all but work with files, data and documents. It works there too; Claude Code for non-coding tasks has examples.

Is Claude Code free? Plans by name

No. As of September 30, 2026, Anthropic’s setup page↗ (opens in a new tab) says Claude Code needs a Pro, Max, Team or Enterprise plan, or a Claude Console account, and that the free claude.ai plan does not include it. Organizations can also run it through Amazon Bedrock, Google Cloud’s Agent Platform or Microsoft Foundry and pay that provider instead.

On Pro and Max, one subscription covers both the Claude apps and Claude Code, and the Help Center↗ (opens in a new tab) says their usage limits are shared: a long coding session and a long chat draw on the same allowance. A Console account is billed by API usage instead. Cloud sessions on the web are available on Pro, Max and Team, and on Enterprise with the right seat type. For current prices, go to Anthropic’s own pricing page; they change, and this page does not quote them.

Where to start

1. Install it with one command: how to install Claude Code covers Mac, Windows and Linux, signing in and checking the install.
2. Spend an hour on one small change: Claude Code for beginners walks through it step by step.
3. Learn the habits that make it trustworthy: Claude Code best practices.
4. Keep the commands cheat sheet open for the first week.
5. When you want the full reference, the map of the Claude Code docs shows what is where.

Where the record of its work lives

Once Claude Code is part of your week, the question changes from what it can do to what it did. The session transcript answers that for one person on one machine. fenbs is a task board where Claude Code is a member: connect it with claude mcp add --transport http fenbs https://fenbs.ai/api/mcp and /mcp, and it reads the board and the rules on the Decisions and rules page through fenbs_get_context, writes each task’s note and plan, moves it from To Do through In Progress to Completed, and records how it was tested. History shows every change under the assistant’s name. fenbs does not do sprints, due dates or assignees; it keeps the list, and who changed what. A task-tracking workflow for Claude Code shows the rhythm.