Everything Claude Code: The Complete A-to-Z Guide

Everything Claude Code offers boils down to one idea: an AI agent that lives in your terminal, reads and edits your whole repository, runs commands, and ships real work. This is the map — install and sign-in, the plan-execute-verify workflow, model switching, sessions, cost, the extension ecosystem, remote control, and the errors you'll hit — with a link to the deep dive on each. Bookmark it as your table of contents to everything Claude Code.

Model lineup and per-token pricing verified 31 July 2026 against the Anthropic pricing page.

We run this site with Claude Code every day, and the tool has grown far past "AI autocomplete." This guide is the whole picture in one place: not a re-explanation of each feature, but a tour of everything Claude Code can do with pointers to the detailed guides when you want to go deeper. It currently runs on the latest Claude models — Opus 5 for the hardest work, Haiku 4.5 for the cheapest — and we re-check every link and claim here roughly every 30 days so the map stays accurate. Start at the top if you're new; jump to a section if you're not.

Everything Claude Code does: the short version

Give it a task in plain English and it explores your codebase, proposes a plan, makes the edits, runs your tests, and commits — all from the terminal, with no copy-pasting between an editor and a chat window.

At its core, Claude Code is Anthropic's agentic coding tool. You give it a task in plain English; it explores your codebase, proposes a plan, makes the edits, runs your tests, and commits — all from the terminal. If you're brand new, our what is Claude Code explainer covers the concept from zero. Anthropic's own overview docs are the canonical reference.

Here's everything Claude Code touches, at a glance:

How do you get started with Claude Code?

Install from npm or the native installer, sign in with your Claude account or an API key, then run claude inside a project directory. It works in five places, but the terminal gets every feature first.

Claude Code installs from npm or a native installer and signs in with your Claude account or an API key. It runs in five places — terminal, VS Code, JetBrains, the desktop app, and the web at claude.ai/code — but the terminal is the reference implementation where every feature lands first. Once installed, keep it current with claude update; a stale build misses new models and commands. If you can't sign in or hit a connection wall, our can't reach Claude error fixes cover it. And if you're wondering whether you can read or fork it, the Claude Code source code is proprietary — not open source.

The core Claude Code workflow

Plan, execute, verify, commit. Start non-trivial work in plan mode so the agent proposes an approach before touching files, approve it, let it edit and run tests, then commit. Sessions save automatically for later resume.

This is the part that makes everything Claude Code does hang together. The loop is: plan, execute, verify, commit. You start non-trivial work in plan mode so the agent proposes an approach before touching files, approve the plan, let it make the edits and run tests, then commit. The whole session — every prompt, edit, and command — is saved automatically as a Claude Code session you can resume with claude --continue.

Choosing and switching models

Sonnet handles the fast everyday bulk, Opus takes the hardest reasoning and large refactors, and Haiku does cheap mechanical work. Switch mid-session with /model, or use opusplan to plan on Opus and execute on Sonnet.

Claude Code runs on the Claude model family: Sonnet for the fast, everyday bulk of work, Opus for the hardest reasoning and large refactors, and Haiku for cheap mechanical tasks. You switch models mid-session with /model, so it's a per-task choice. The opusplan mode even uses Opus to plan and Sonnet to execute — the best of both.

What does Claude Code cost?

Two routes. A Claude subscription bundles usage at a flat fee from $20 a month, or an API key bills per token. Daily coders usually come out ahead on a subscription; automation and spiky use favour the API.

Cost is one of the first things people ask about everything Claude Code involves. Two routes: a Claude subscription (Pro at $20/month, Max at $100 or $200) bundles usage at a flat fee, or an API key bills per token — Haiku 4.5 at $1/$5, Sonnet 5 at $3/$15 (discounted to $2/$10 through 31 August 2026), and Opus 5 at $5/$25 per million input/output tokens, all listed on the Anthropic pricing page. For most individual developers who code daily, a subscription is cheaper; the API wins for automation and spiky use. The full breakdown — including whether Claude Code is free and the per-day numbers — is in our Claude Code cost guide. Heavy daily coders should also weigh Claude Code on the Max plan, and if cost or lock-in is a dealbreaker, open-source alternatives exist — see OpenCode vs Claude Code.

Extending Claude Code: skills, MCP, plugins, and more

Five mechanisms: skills encode reusable procedures, MCP servers connect external tools and data, hooks fire shell commands at lifecycle events, subagents run focused parallel work, and plugins install whole capabilities in one command.

This is where everything Claude Code can do expands to fit your stack. Five mechanisms matter:

• Skills encode reusable procedures the agent loads on demand.
• MCP servers connect Claude Code to your databases, APIs, and tools — even creative apps like Blender.
• Hooks fire shell commands at lifecycle events to automate your workflow.
• Subagents spawn focused helpers with their own context for parallel work.
• Plugins drop whole capabilities in with one command — from Anthropic's official ones to community workflow systems like Get Shit Done.

Together they turn a general coding agent into one that knows your team's conventions and connects to the systems your code actually touches.

Claude Code beyond the terminal

The desktop app adds a GUI, Remote Control lets you steer a local session from your phone or browser, and the web version runs each session in an Anthropic-managed cloud sandbox for fire-and-forget tasks.

Everything Claude Code does isn't locked to one window. The desktop app gives it a GUI, and Remote Control lets you steer a local session from your phone or browser while it keeps running on your machine — start a refactor at your desk, approve steps from the couch. The web version at claude.ai/code runs each session in an Anthropic-managed cloud sandbox for fire-and-forget tasks.

Common Claude Code errors (and where to fix them)

Check Anthropic's status page first, then update the CLI, then start a fresh session. That trio clears the large majority of failures, including the connection error people most often mistake for something they broke.

Even a tool this capable throws the occasional wall. The good news: almost every error has a known fix. The most common is the can't reach Claude error, usually a connection or status issue rather than anything you broke. When something misbehaves, check status first, then update, then start a fresh session — that trio clears the majority of problems, and the specific fix guides on this site cover the rest.

Everything Claude Code changes about how you code

The work moves from typing to directing. You review far more than you write, whole-repo tasks become single prompts, your conventions get encoded once, and verification becomes part of the loop rather than a follow-up.

The reason everything Claude Code does adds up to more than a faster autocomplete is the shift in where the work happens. With an inline assistant, you still hold the whole task in your head — you type, it suggests, you accept. With Claude Code you delegate and supervise: you describe the outcome, review the plan, and approve the steps while the agent holds the context.

That changes the day-to-day in concrete ways:

• You review more than you type. Most of your input becomes "yes, continue" or "no, do it this way" rather than writing every line — which is exactly why plan mode and clear approval gates matter so much.
• Whole-repo tasks become normal. Renaming a concept across forty files, wiring a new endpoint end to end, or migrating a dependency turn into single prompts instead of afternoons.
• Your knowledge gets encoded once. Skills, a good CLAUDE.md, and MCP connections mean the agent stops needing the same explanations every morning.
• Verification is part of the loop. The agent runs your tests and checks its own work, so "done" means "done and checked," not "here's a draft."

Anthropic pitches Claude Code for exactly this kind of agentic, whole-task development on its product page, and in practice that's where it earns its place: not the one-liner you could type faster yourself, but the multi-step work where planning, context, and verification actually pay off.

It's worth learning if you write code regularly and want to move up a level — from writing every line to directing an agent that writes and verifies them. The curve is short: install it, run one real task in plan mode, and you'll feel the difference within an hour. Everything Claude Code offers builds on that single habit of delegating and reviewing instead of typing and accepting.