A dark industrial terminal glowing with a red screen, representing a command-line coding agent that works inside your shell.
Claude Code lives where developers already work: the terminal and the IDE, talking straight to the model with no extra backend in between.

Claude Code is Anthropic’s agentic coding tool. It runs in your terminal, reads your whole project, and edits files, runs commands, and works through multi-step tasks while you watch. It solves the problem of context-switching between a chat window and your editor: instead of copying snippets back and forth, you hand Claude Code a task in plain language and it works directly in your repository. It is the same agentic engine that powers Claude Cowork for knowledge work, pointed at code.

Where it lives

Claude Code is primarily a program you run on your own machine, and it connects straight to the Claude model API with no backend server or remote code index in between. It can also run as a cloud session on Anthropic-managed infrastructure (or your organization’s self-hosted environment), started from the browser, phone, desktop app, or terminal; cloud sessions are available on Pro, Max, and Team plans and for eligible Enterprise seats.

Where you run it
Terminal VS Code and forks Cursor JetBrains (IntelliJ, PyCharm) GitHub Actions
The agent
Claude Code Plans a task, edits files, runs commands, checks the result
The model
Claude Opus 5.5 Claude Sonnet 5 Claude Fable 5.1 Reached directly over the API, no remote index
Your environment
Local files and git Shell commands MCP tools and servers

Install it

Anthropic’s recommended install is the native installer, which updates itself in the background. Homebrew (brew install --cask claude-code), WinGet (winget install Anthropic.ClaudeCode), Linux package managers, and npm (npm install -g @anthropic-ai/claude-code) are also supported. Once installed, it runs inside any project directory.

bash
# Install once (macOS, Linux, WSL)
curl -fsSL https://claude.ai/install.sh | bash

# Start an interactive session in your project
cd my-project
claude

What it is for

Once a session is open, you describe work in plain language and Claude Code carries it out across your files.

text
> explain how authentication flows through this codebase
> fix the failing test in tests/login.spec.ts, then run the suite
> add a rate limiter to the /api/upload route and update the docs

For automation and scripting, the headless mode runs a one-shot task and prints the result, which is useful in CI and shell pipelines.

bash
# Headless: run a single task and print the output
claude -p "summarize the changes in the last 5 commits"

A typical task moves through the same loop every time, which is what makes it an agentic loop rather than a single answer.

Step 1DescribeYou hand Claude Code a task in plain language inside your project.
→
Step 2PlanIt reads the relevant files and breaks the work into concrete steps.
→
Step 3ActIt edits files and runs commands, asking before risky actions.
→
Step 4CheckIt runs tests or builds, reads the output, and fixes what broke.

Recent changes

  • Projects as multi-agent cloud threads (17 September 2026). Anthropic relaunched Projects in Claude Code as a way to run several agents under one goal. A coordinator scopes the request and delegates it to parallel threads; each thread is a full Claude Code cloud session working on its own git branch and copy of the repo, and can split its work further with subagents. Where threads touch the same code, the overlap surfaces as an ordinary merge conflict. Threads run in the cloud at launch, with local tools and code support described as coming soon. It is a beta for select Pro and Max subscribers (Team and Enterprise later), and Anthropic warns that because every thread is a full session, projects reach usage limits faster. This is distinct from the older claude.ai Projects workspaces. Unlike the local terminal flow shown above, this mode runs on Anthropic’s infrastructure rather than your machine. See Claude Code relaunches Projects as coordinated parallel cloud threads .

  • Claude Opus 5.5 (22 September 2026). The new Opus model is available in Claude Code, including fast mode (up to 2.5x faster output at $8/$40 per million tokens on API billing). See Claude Opus 5.5 launches at $4/$20, with thinking always on and Claude by Anthropic for the full lineup and pricing.

  • Plugin4Shell (disclosed 17-18 September 2026). AIR Security disclosed a zero-click remote-code-execution flaw affecting Claude Code, OpenAI Codex, GitHub Copilot and Gemini CLI: an attacker who controls a plugin repository could create a branch named like the marketplace’s pinned 40-character commit SHA, so the agent’s background plugin update checked out malicious code while appearing to honour the pin. According to AIR, Anthropic fixed it in Claude Code 2.1.179 (released 16 June 2026), and no in-the-wild exploitation is known. If you run an older build or install plugins from Bitbucket, GitLab or self-hosted marketplaces, update Claude Code and review which plugins auto-update. See Plugin4Shell: a zero-click plugin flaw in coding agents . Anthropic’s own advisory followed on 9 October 2026 as GHSA-pq7j-f95f-qfcv / CVE-2026-86063 (High, CVSS 4.0 7.7 in Anthropic’s advisory; the CVE record on NVD gives 7.5), and it says Claude Desktop’s plugin install flow was affected too.

  • Security advisories disclosed in October 2026, all fixed in earlier releases. Anthropic published three advisories that month. The minimum safe versions are the fixed versions they name:

    AdvisoryWhatSeverity (Anthropic)Fixed in
    GHSA-pq7j-f95f-qfcv / CVE-2026-86063 (9 Oct)Plugin commit SHA pin bypass, Claude Code and Claude Desktop (CWE-494, CWE-829)High, CVSS 4.0 7.7Claude Code 2.1.179
    GHSA-4p2j-qj3c-6x84 / CVE-2026-88066 (13 Oct)Claude Desktop on Windows ran git.exe without an absolute path, so a git.exe planted in a project directory ran when the project was opened (CWE-427)High, CVSS 4.0 7.5Claude Desktop 1.8089.0
    GHSA-5j29-h97v-84ch / CVE-2026-103435 (5 Oct)Write-time symlink following (TOCTOU): a path checked at permission time was re-resolved at write time, so an attacker with write access to a shared workspace could redirect a write outside the project (CWE-22, CWE-61, CWE-367)High, CVSS 4.0 7.7Claude Code 2.1.129

    Each advisory says users on standard auto-update “have received this fix already”. If you pin versions or update by hand, check that you are at or above these. The October 2026 agent security roundup covers them alongside the MCP SDK and framework advisories.

Which subscription you need

Claude Code is covered by a normal Claude subscription, so you do not pay separately for the chat apps and the coding agent.

  • Pro (about 19 EUR per month, listed at 20 US dollars): includes Claude Code in the terminal and in supported IDEs, alongside the web, desktop, and mobile apps on one subscription.
  • Max (about 92 or 185 EUR per month, listed at 100 or 200 US dollars): the same access with 5x or 20x the usage and priority on new models.
  • Team Premium and Enterprise: Claude Code for organizations, with admin controls and billing.
  • API: pay-per-token billing for automation. Since 15 June 2026, programmatic use (the Agent SDK, the claude -p headless command, the GitHub Actions integration, and third-party apps) draws from a separate monthly Agent SDK credit at standard API rates, rather than your interactive subscription pool.

The Claude product family

Claude Code is one of several products built on the same models. They differ mainly in where they run and what they produce.

Where it livesWhat it is forPlan needed
Claude CodeTerminal and IDEsEditing, running, and shipping codePro, Max, Team Premium, or API
Claude Designclaude.ai (Anthropic Labs)Designing UI and documents as HTMLPro, Max, Team, Enterprise
Claude CoworkClaude app (desktop, and since September 2026 web and mobile)Autonomous multi-step knowledge workPaid plans (Pro and Max first for the merged app)
Claude apps and APIWeb, mobile, desktop, APIChat, analysis, building on the modelFree and up

When not to use it

  • You do not write or run code. For document and file work without a terminal, Claude Cowork fits better.
  • You want a chat answer, not file edits. The Claude apps are the lighter choice for questions and drafting.
  • You need a visual prototype. Claude Design produces editable interface and document layouts; Claude Code produces working code.
  • Your budget is strict and usage is heavy. Watch the plan limits, since large agent runs consume usage quickly and programmatic runs bill at API rates.

Further reading