Claude Code Complete Guide (2026 Edition)

This Claude Code guide in one paragraph: install Claude Code, run claude in a project, log in with a paid Claude plan or a Console account, describe the outcome you want, and let Claude explore, plan, edit and run your tests before you review the diff. The rest of 2026's feature set, from CLAUDE.md to subagents, hooks, skills and MCP, exists to make that loop cheaper, safer and repeatable.

Checked October 2026 against Anthropic's Claude Code docs: overview, setup, best practices, permission modes, subagents, hooks, skills, MCP, memory, model configuration and costs, plus the Claude pricing page. Claude Code ships often; flags and defaults below can move between versions.

This is the annual refresh of our Claude Code guide. It replaces the May 2026 version, which predated several of these features. If you only need the install steps, our step-by-step install guide is shorter. If MCP is new to you, read our Model Context Protocol guide alongside this one; it is the hub for everything Claude Code connects to.

What Claude Code is, and where it runs

Claude Code is an agentic coding tool, not a chat window with a code box. It reads your files, runs shell commands, edits code across many files, stages commits and opens pull requests. You can run the same agent in five places:

• Terminal CLI: the full-featured version, and the one you can script with claude -p in CI.
• VS Code and Cursor: an extension with inline diffs, @-mentions and plan review. Our Claude Code in VS Code setup guide covers the settings.
• JetBrains IDEs: a plugin that needs the CLI installed separately.
• Desktop app: the Code tab in the Claude app, with side-by-side sessions and scheduled tasks. It needs a paid subscription.
• Web: claude.ai/code, for long tasks on repositories you do not have locally, including from the mobile app.

The mental model that makes Claude Code click

Two ideas explain almost every best practice in Anthropic's docs. First, Claude Code runs a loop: it gathers context, plans, acts, checks the result and repeats until the work looks done. Second, everything in that loop, every file it reads and every command output, lands in one context window, and Anthropic says performance degrades as that window fills. The context window is the budget you are managing.

So the two habits that matter most follow directly. Give Claude a check it can run, because without one "looks done" is the only stop signal and you become the test suite. And keep the context clean: one task per session, /clear between unrelated tasks, and side research pushed into subagents so only the summary comes back.

Claude Code tutorial: what does a first session look like?

The setup docs list macOS 13 or later, Windows 10 1809 or later, Ubuntu 20.04, Debian 10 or Alpine 3.19, and 4 GB of RAM. On native Windows, Git for Windows is recommended so Claude can use Bash; without it Claude Code falls back to PowerShell. The npm package still exists and needs Node.js 22 or later, but it installs the same native binary. Do not install it with sudo.

How do you use Claude Code day to day?

Prompts that name the symptom, the location and what "done" looks like beat short requests every time. Compare "fix the login bug" with "users report that login fails after session timeout; check the token refresh in src/auth/, write a failing test that reproduces it, then fix it." The second gives Claude a place to look and a check to pass.

• Point at files with @ instead of describing where code lives.
• Press Esc to stop Claude mid-action and redirect it; context is kept.
• Press Esc twice or run /rewind to restore earlier code, conversation or both. Claude snapshots files before each change.
• Run /clear between unrelated tasks, and after correcting Claude twice on the same issue. A fresh session with a better prompt usually wins.
• Run /compact Focus on the API changes when you need to keep going but want the history condensed.
• Skip planning for one-line changes. Anthropic's rule of thumb: if you could describe the diff in one sentence, skip the plan.

Permission modes

Permission modes decide what Claude may do without asking. Switch them mid-session with Shift+Tab or start with claude --permission-mode plan. Recent versions start interactive sessions in auto mode where it is available, which has a separate classifier review actions in the background.

Claude Code guide to the 2026 features: CLAUDE.md, skills, subagents, hooks and MCP

Each of these answers the same question differently: where should an instruction live so Claude follows it at the right moment without wasting context?

CLAUDE.md: the always-on instructions

Claude Code reads CLAUDE.md at the start of every session, from your working directory and every directory above it, plus ~/.claude/CLAUDE.md for personal preferences and a gitignored CLAUDE.local.md for private project notes. If your repo already has an AGENTS.md, Claude Code can read that instead. The docs recommend keeping each file under 200 lines: put in what Claude cannot work out from the code (commands, environment quirks, rules) and leave out file-by-file descriptions it can read itself.

Skills

A skill is a folder with a SKILL.md file: YAML frontmatter that says when to use it, then the instructions. Unlike CLAUDE.md, the body loads only when the skill is used, so long material costs almost nothing until needed. Custom slash commands have been merged into skills, so .claude/commands/deploy.md and .claude/skills/deploy/SKILL.md both create /deploy. Claude Code also bundles skills such as /code-review, /debug and /verify, and skills follow the open Agent Skills standard.

Subagents

A subagent runs in its own context window with its own system prompt, tools and model, and returns only a summary. Claude Code includes built-in ones, such as Explore, a fast read-only agent for searching a codebase, and a general-purpose agent. You add your own as Markdown files in .claude/agents/:

---
name: code-reviewer
description: Reviews code for quality and best practices
tools: Read, Glob, Grep
model: sonnet
---

You are a code reviewer. When invoked, analyze the code and provide
specific, actionable feedback on quality, security, and best practices.

That example is from Anthropic's docs. Only name and description are required, and Claude uses the description to decide when to delegate, so keep it short and specific. Subagent requests count toward the same usage limits as your main session; the saving is context, not quota.

Hooks

Hooks are shell commands Claude Code runs at fixed points: PreToolUse (which can block a tool call), PostToolUse, SessionStart, Stop, Notification and more. They are deterministic: the action always happens, instead of depending on the model remembering. Anthropic's example formats every file Claude edits:

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Edit|Write",
        "hooks": [
          { "type": "command", "command": "jq -r '.tool_input.file_path' | xargs npx prettier --write" }
        ]
      }
    ]
  }
}

Put it in .claude/settings.json and check it with /hooks. A Stop hook that runs your test script and blocks the turn until it passes is the strongest version of "verify before you say done".

MCP servers

MCP connects Claude Code to tools outside your repo: issue trackers, docs, databases, design files. Add a remote server with claude mcp add --transport http notion https://mcp.notion.com/mcp, or commit a .mcp.json so the whole team shares it. Claude Code uses MCP tool search by default, so tool definitions load when Claude looks for them rather than all at once, but every server is still code you trust with access, so connect what a project needs rather than everything you use. Our MCP guide explains servers, clients and transports in depth.

Scripts, CI and parallel sessions

Claude Code follows the Unix habit of doing one job and taking input from a pipe. claude -p "prompt" runs a single non-interactive task and prints the result, so you can pipe logs into it or call it from a script. Anthropic's examples include piping the last 200 lines of a log for anomalies and passing a list of changed files from git diff for a security review.

In CI, pair it with an exact allowlist so nothing unexpected runs: claude -p "run the test suite" --permission-mode dontAsk --allowedTools "Bash(npm test)" "Read". Anything that would prompt is denied instead. For code review and issue triage there are official GitHub Actions and GitLab CI/CD integrations. Keep bypassPermissions (the --dangerously-skip-permissions flag) for containers or VMs you can throw away; the docs say so plainly.

For more than one task at a time, you have three levels. Subagents split work inside one session. Background agents run several full sessions you can watch from one screen. Agent teams let a lead session spawn and supervise others. Give parallel sessions their own git worktree so they do not edit the same files; subagents can do this with isolation: worktree in their frontmatter. More sessions means more usage, so start with one and add a second only when the first is waiting on you.

Claude Code pricing, models and costs

Anthropic's setup docs say Claude Code needs a Pro, Max, Team, Enterprise or Console account; the free plan does not include it. Checked October 2026 on the pricing page, Pro is $17 a month billed annually ($200 up front) or $20 monthly, and Max starts at $100 a month with 5x or 20x Pro's usage. With a Console account you pay per token instead: Claude Opus 5.5 lists $4 per million input tokens and $20 per million output, Sonnet 5.5 $2 and $10, and Haiku 5.5 $0.10 and $0.50 for prompts up to 100K tokens. Our LLM API pricing comparison sets those against OpenAI and Gemini.

Pick models by alias rather than version: sonnet for daily coding, opus for complex reasoning, haiku for simple tasks, opusplan to plan with Opus and execute with Sonnet, and fable or best for the hardest long-running work where your plan includes Fable. On the Anthropic API the aliases currently resolve to Opus 5.5, Sonnet 5.5 and Haiku 5.5. A cheap trick from the subagent docs: define your own Explore subagent with model: haiku to run codebase searches on the lowest-cost model.

Claude Code vs Cursor and other AI coding tools

AI editors such as Cursor and Windsurf (now Devin Desktop) started from inline completion and grew agents; Claude Code started as an agent and grew editor integrations. Claude Code's own extension runs inside Cursor, so it is not either-or. Choose Claude Code when you want one configuration (CLAUDE.md, hooks, subagents, MCP) that works in your terminal, CI, the desktop app and the browser. Our Cursor vs Windsurf comparison and Cursor guide cover the editor side, and our vibe coding guide covers building with these tools when you are not a full-time developer.

Your first week with Claude Code

Claude Code is one tool in building a real product. Our AI SaaS Builder program uses it, together with the Claude API, Supabase, Next.js and Stripe, to take an idea to a paid product.

Related reads

• Model Context Protocol (MCP) explained
• Claude Code in VS Code: setup and settings
• How to get a Claude API key
• Claude API pricing tutorial
• How to get an OpenAI API key
• Claude vs GPT-4 comparison
• Claude vs Gemini comparison

Claude Code: FAQ