A genuinely capable terminal-first coding agent — best for developers who already live in a shell, not for anyone who wants a GUI to hold their hand.

What it actually is

Claude Code is Anthropic’s agentic coding tool: it reads your codebase, edits files, runs shell commands, and integrates with your existing development tools, rather than being a chat window you paste code into. Anthropic’s own framing is blunt about this — it’s “available in your terminal, IDE, desktop app, and browser,” which tells you the real center of gravity is the terminal; the other surfaces are conveniences layered on top (Claude Code overview).

It shipped in February 2025 as an agentic CLI tool, went generally available in May 2025 alongside the Claude 4 model family, picked up a Chrome extension in August 2025, and got a full web version with sandboxing in October 2025. A more GUI-oriented sibling aimed at non-technical users, Claude Cowork, arrived in January 2026 — worth knowing about if a terminal genuinely isn’t for you (timeline via Wikipedia).

What you get

Beyond the core “reads and edits your code” loop, the current feature set is broad: subagents that run specialized tasks in their own context window, hooks that fire deterministic shell commands at lifecycle events (before a tool runs, after it runs, when a session stops), official MCP support for wiring in external tools and data sources, persistent project memory via CLAUDE.md, background and parallel agent runs, an Agent SDK for building custom agents on the same engine, cloud-scheduled “Routines,” and CI integrations for GitHub Actions and GitLab (Claude Code overview).

As of late June 2026, Claude Code’s default model is Sonnet 5, with a native 1-million-token context window and promotional pricing through the end of August 2026; “Claude in Chrome” — an in-browser agent mode — reached general availability the following week. Both of those are recent enough that the details are worth re-checking against the official changelog before you plan around them, since this kind of thing shifts fast.

Pricing, as of writing

Claude Code itself isn’t a separate subscription — it’s bundled into Anthropic’s existing plans, and only counted separately if you connect it to a metered API key. At the time of writing: Pro runs $17/month billed annually (or $20/month billed monthly); Max starts at $100/month; Team’s Standard seat is $20/month annually (or $25/month monthly) and explicitly includes both Claude Code and Claude Cowork; Team Premium is $100/month annually; Enterprise is $20/seat plus usage billed at API rates. All plans carry usage limits that Anthropic doesn’t itemize on the pricing page itself. Check claude.com/pricing directly before you commit — plan pricing and limits are exactly the kind of thing that changes without a rewrite of every review site that quoted the old numbers.

If you’re running it against a raw Anthropic API key instead of a subscription plan, you’re paying metered token rates, which can get expensive fast on a big codebase — Anthropic’s own enterprise cost data (self-reported, not an independent benchmark) puts average cost around $13 per developer per active day, with 90% of users staying under $30/day (source). Take that as an order-of-magnitude signal, not a quote for your workflow.

The honest caveats

• It wants a terminal. The IDE extensions (VS Code, and a JetBrains plugin currently in beta) and the desktop/web apps soften this, but the tool’s mental model — plan, edit, run, diff, repeat — is a CLI workflow wearing different skins. If reading a unified diff in a terminal makes you anxious, this isn’t the product that fixes that for you.
• API billing can surprise you. If you skip the subscription plans and wire it to a raw API key on a large repo, cost is usage-based and unbossed by a flat monthly cap. Know which mode you’re in before you let it loose on a big refactor.
• The feature surface moves fast. Subagents, hooks, and Routines are all genuinely useful, but they’re also the kind of features that get renamed or reshaped between releases — the changelog is worth a skim before you build a workflow around any one of them.

Who it’s actually for

Good fit: developers who already work in a terminal daily, teams that want an agent wired into CI (GitHub Actions/GitLab), and anyone comfortable reviewing a diff before it lands. Weaker fit: developers who want a fully graphical, click-driven experience with no shell in sight — Claude Cowork exists for exactly that gap, but it’s a different product with a different learning curve, not Claude Code with a coat of paint.

If you’re weighing it specifically against GitHub Copilot’s higher tiers — which, confusingly, now bundle Claude Code access as a third-party agent — see our Claude Code vs GitHub Copilot comparison for the full pricing table and where the two tools’ architectures actually diverge. For a practical setup once you’ve picked it, see getting real work out of AI agents.