What Is Claude Code? Anthropic's AI Coding Agent Explained

Claude Code is Anthropic's agentic coding tool: a command-line program (with companion extensions for VS Code and JetBrains, a desktop app, and a browser version) that reads your codebase, edits files, runs shell commands and tests, and carries out multi-step engineering tasks from a plain-language instruction. It is not a separate product you buy stand-alone — on Anthropic's Claude.ai subscription plans it is a feature bundled into Pro and above, metered against the same usage pool as your chats, while on an Anthropic Console (API) account it is billed purely by token consumption, the same way any other call to the Claude API is. Which path makes sense depends on whether you want predictable monthly billing or pay-as-you-go metering, and the rest of this explainer breaks down both, plus the models it runs on, what it can actually do, and how to install it.

What is Claude Code?

Claude Code is Anthropic's agentic coding tool. Anthropic describes it as a tool that "reads your codebase, edits files, runs commands, and integrates with your development tools," available in a terminal, an IDE, a desktop app, and a browser, according to the official Claude Code documentation. The core distinction from a traditional code-completion assistant is the word "agentic": instead of only suggesting the next line as you type, Claude Code can be given an open-ended instruction — "write tests for the auth module, run them, and fix any failures" is the documentation's own example — and it will plan the work, search the codebase for relevant files using agentic search rather than requiring you to manually attach context, make the edits across however many files are involved, run the resulting tests or build, and report back.

It ships from Anthropic as a single underlying engine that surfaces in several places: a full-featured terminal CLI, extensions for VS Code and JetBrains IDEs, a standalone desktop app for macOS and Windows, a browser-based version at claude.ai/code, and integrations with GitHub Actions, GitLab CI/CD, and Slack (where tagging @Claude with a bug report can return a pull request). Every surface shares the same project settings, memory files, and Model Context Protocol server connections, so a session started on the web can later be picked up from the terminal.

How much does Claude Code cost?

This is the detail people search for most, and it genuinely depends on how you access Claude. According to Anthropic's current pricing page, there are two distinct billing models:

Through a Claude.ai subscription — Claude Code is a plan feature, not a separate line item:

• Free ($0): does not include Claude Code.
• Pro ($17/month billed annually at $200 up front, or $20/month billed monthly): includes Claude Code, alongside more usage, Claude in Chrome, Claude Design/Slides/Docs, and access to more Claude models.
• Max (from $100/month, with a choice of 5x or 20x the usage of Pro): includes Claude Code with higher output limits and priority access at busy times.
• Team: a Standard seat ($20/month billed annually, $25/month billed monthly) and a Premium seat ($100/month billed annually, $125/month billed monthly) both include Claude Code.

On every one of these plans, Claude Code usage is not billed separately — it draws from the same rolling usage allowance as regular Claude chat, so a long coding session and a long chat conversation compete for the same pool, as Anthropic's cost-management documentation explains.

Through the Anthropic API (a Console account): Claude Code instead charges by token, at the same per-model rates as any other API call, billed to a workspace that the Console automatically creates the first time you authenticate Claude Code that way. There is no flat monthly fee on this path — cost scales with how much context and output each task uses. Anthropic's documentation gives a rough enterprise benchmark of around $13 per developer per active day, or roughly $150–250 per developer per month, though it stresses that actual spend "varies widely based on model selection, codebase size, and usage patterns."

Subscription vs. pay-per-token: which should you pick?

The table below lines up the two paths so the trade-off is explicit.

Anthropic notes that Claude Code computes the dollar figure shown in its own /usage command from token counts at list price; for authoritative billing on an API account, the Claude Console's usage page is the source of truth.

Which Claude models does Claude Code run on?

Claude Code is a harness, not a model — it can be pointed at any current Anthropic model, selected with the in-session /model command, the claude --model flag, or the ANTHROPIC_MODEL environment variable. Per Anthropic's model configuration documentation, the supported lineup spans Anthropic's Opus, Sonnet, Haiku, and Fable model families — Opus is positioned for complex reasoning and architectural decisions, Sonnet for everyday coding and agent work at a lower cost, and Haiku for fast, simple subtasks (it's a common choice for subagents). On Pro, Max, Team, Enterprise, and standard API accounts, Opus is the default model unless you change it. Convenience aliases such as opus, sonnet, and haiku always resolve to the latest version in that family, and an opusplan alias uses Opus for planning while automatically switching to Sonnet for execution to control cost.

Model choice is also the single biggest lever on cost: Anthropic's own cost-reduction guidance says plainly that "Sonnet handles most coding tasks well and costs less than Opus," and recommends reserving Opus for genuinely hard architectural or multi-step reasoning problems.

Key features: what Claude Code can actually do

Beyond single-prompt code generation, the documentation highlights several capabilities that define Claude Code as an agent rather than an autocomplete tool:

• Multi-step, multi-file agentic coding. Claude Code plans an approach, asks clarifying questions when needed, edits across as many files as the task requires, and verifies its own work by running tests or builds.
• Git and GitHub/GitLab integration. It stages changes, writes commit messages, creates branches, and opens pull requests directly, and can automate PR review and issue triage inside GitHub Actions or GitLab CI/CD.
• MCP server support. Through the open Model Context Protocol, Claude Code can connect to external tools and data — reading design docs in Google Drive, updating Jira tickets, or pulling from Slack — without custom integration code.
• Subagents and the Agent SDK. A session can spawn multiple Claude Code agents to work on different parts of a task in parallel, coordinated by a lead agent; for fully custom automation, Anthropic also ships an Agent SDK that exposes Claude Code's harness and tools for building bespoke agents.
• Hooks. Shell commands can be configured to run automatically before or after specific Claude Code actions — for example, auto-formatting a file right after it's edited, or running a linter before a commit is created.
• Skills and persistent memory. A CLAUDE.md file at the project root is read at the start of every session to set coding standards and conventions, and reusable workflows can be packaged as "skills" invoked with slash commands like /review-pr.
• CLI composability. Running non-interactively with the -p flag lets Claude Code be piped into or chained with other command-line tools — for example, tail -200 app.log | claude -p "Slack me if you see any anomalies".
• Scheduled and remote work. Routines can run recurring tasks in the cloud even when a machine is off, and a session can be started on one surface (terminal, web, mobile) and continued on another.

IDE extensions and other ways to run it

Claude Code is not limited to a terminal window. The VS Code extension adds inline diffs, @-mentions, and plan review directly in the editor and can be installed by searching "Claude Code" in the Extensions view. A JetBrains plugin (covering IntelliJ IDEA, PyCharm, WebStorm, and other JetBrains IDEs) is available from the JetBrains Marketplace, though it still requires the Claude Code CLI to be installed separately. A standalone desktop app for macOS and Windows bundles the CLI and adds visual diff review, side-by-side sessions, and scheduled tasks, while the web version at claude.ai/code runs sessions entirely in the browser with no local setup — useful for kicking off a long-running task and checking back later, or working on a repository you don't have cloned locally.

How to install Claude Code

The official GitHub repository and documentation list several installation routes. The native installer is recommended and updates itself automatically in the background:

• macOS, Linux, or WSL: curl -fsSL https://claude.ai/install.sh | bash
• Windows PowerShell: irm https://claude.ai/install.ps1 | iex
• Windows CMD: curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd
• Homebrew (macOS/Linux): brew install --cask claude-code (does not auto-update; run brew upgrade periodically)
• WinGet (Windows): winget install Anthropic.ClaudeCode
• npm: npm install -g @anthropic-ai/claude-code — listed in the GitHub README as a deprecated installation path in favor of the native installer

Linux users can also install through apt, dnf, or apk package managers on supported distributions. Once installed, running claude --version confirms a working setup, and running claude from inside any project directory starts a session — the first run prompts you to log in with a Claude.ai account, or, if the ANTHROPIC_API_KEY environment variable is already set, asks you to approve using that key instead.

Realistic use cases and limitations

Anthropic's own documentation frames Claude Code's best use cases as the tedious, well-specified work developers tend to put off: writing tests for untested code, fixing lint errors across a project, resolving merge conflicts, updating dependencies, and drafting release notes, plus larger jobs like building a described feature or tracing a bug report to its root cause across a codebase. It's positioned to compete in the same general category as other AI-assisted coding tools — including editor-native agents and GitHub Copilot's own agentic features, and IDE-first competitors like Cursor and Windsurf — though Claude Code's own distinguishing pitch is running the same agent consistently across terminal, IDE, desktop, web, and CI rather than being tied to one editor.

The limitations worth keeping in mind are practical ones. Claude Code asks for permission before running commands or editing files by default (permission modes can be configured, including a plan-first mode that proposes an approach before touching anything), so it is not meant to run fully unattended on sensitive systems without review. Cost and token usage scale with how much context a session carries — a long-running session that's never cleared, or defaulting to the most capable model for simple tasks, is the most common cause of unexpectedly high bills, according to Anthropic's cost guidance. And because it is a generative agent, its output still needs the same code review, tests, and verification any human-written change would get before merging.

The bottom line

Claude Code is Anthropic's answer to "what if an AI could actually do the coding task, not just suggest the next line" — a terminal-first agent, also available in VS Code, JetBrains, a desktop app, and a browser, that reads a codebase, makes multi-file changes, runs commands and tests, and opens pull requests. Cost is the one detail worth rechecking before you commit: on a Claude.ai subscription it's bundled into Pro ($17–20/month) and above with no separate per-token charge, while on the Anthropic API it's metered like any other model usage, at rates such as $2/$10 per million input/output tokens for Sonnet 5.5. Anthropic updates both the product and its pricing fairly often, so for an exact current number, Anthropic's own pricing page and cost-management docs remain the sources to check directly before budgeting.

Written byFelix Moreau

Felix Moreau writes Pandromeda's software coverage and how-to guides. He covers Windows, macOS and Linux updates, the apps people rely on, emulators and developer tools, and turns official documentation into clear, numbered steps that work on the current version.

More from Felix MoreauAll Software & Guides coverage