Claude Code Docs: The 2026 Quick Reference

Claude Code docs, in one place

The official Claude Code docs live at docs.anthropic.com (Anthropic also publishes these under docs.claude.com). Those pages are the source of truth for installation, configuration, commands, hooks, MCP, and the SDK. This page is the fast companion to them: where to look, the core concepts a new user actually needs, and a cheatsheet of the most-used commands, CLI flags, and config files you will reach for every day.

Claude Code is Anthropic's terminal-native AI coding agent. You install it with npm, run claude in any project, and it reads your codebase, runs commands, and executes multi-step tasks. The docs are thorough but reference-dense, so the goal here is to give you a map and a cheatsheet you can scan in a minute, then send you to the official Claude Code docs for the deep detail.

If you are brand new, start with our introduction to Claude Code, then come back here to bookmark the commands and config.

Where the official docs live

You want	Official docs section
Install and first run	docs.anthropic.com (search "Claude Code quickstart")
Every CLI flag	docs.anthropic.com (search "CLI reference")
Slash commands	docs.anthropic.com (search "Claude Code commands")
Memory and CLAUDE.md	docs.anthropic.com (search "memory")
Settings and permissions	docs.anthropic.com (search "settings")
Hooks	docs.anthropic.com (search "hooks")
MCP servers	docs.anthropic.com (search "MCP")
Agent SDK	docs.anthropic.com (search "agent SDK")
Models and API pricing	docs.claude.com / docs.anthropic.com

The exact page structure shifts as Anthropic reorganises the docs, so when in doubt, open docs.anthropic.com and use its search. Treat the official Claude Code docs as canonical and this page as the quick reference on top.

Core concepts to learn first

The docs cover a lot, but you only need a handful of concepts to be productive. Learn these and the rest of the documentation makes sense.

The agentic loop

Claude Code does not just generate code and hand it back. It reads your project, plans the work, runs commands, checks the result, and fixes what broke. You give it a goal, it handles the loop. This is the single idea that separates it from inline autocomplete tools, and it is why terminal access matters: Claude can actually run your tests and read the errors.

CLAUDE.md

CLAUDE.md is a plain markdown file in your project root that Claude reads at the start of every session. It holds your tech stack, conventions, key commands, and constraints. It is the highest-leverage configuration you can write, and it cascades: a global ~/.claude/CLAUDE.md sets personal defaults, a project ./CLAUDE.md overrides it, and directory-level files override that. Run /init to generate a starter file and /memory to edit it. Our CLAUDE.md guide covers the structure in depth.

Skills

Skills are reusable procedures stored as markdown in .claude/skills/. When a task matches a skill's description, Claude loads it and follows the steps, so you get consistent results instead of ad-hoc improvisation. Skills are where most teams encode their real workflows. See Claude Code Skills Explained.

Hooks

Hooks are deterministic scripts wired to lifecycle events in .claude/settings.json. Unlike skills, which are advisory, hooks are hard constraints Claude cannot skip. Use them to run linting before a commit or block writes that contain secrets. The full pattern is in our Claude Code hooks guide.

MCP servers

MCP (Model Context Protocol) is an open standard from Anthropic that lets Claude Code talk to external systems: databases, GitHub, browsers, and more. You add a server with the claude mcp add command, and its tools become available in your session. Start with the few you actually use, since each one adds to context. Our MCP servers guide walks through setup.

Settings and permissions

Configuration is merged across layers, lowest to highest precedence: your user ~/.claude/settings.json, the project .claude/settings.json, a local .claude/settings.local.json for personal overrides, CLI flags, and any enterprise-managed policy on top. Permissions control which tools Claude can run without asking, which is what stops the constant approval prompts on trusted commands.

The command cheatsheet

These are the built-in slash commands you will use most. Type / in a session to see the full list, or /help to list everything including your custom commands. The set evolves between releases, so the official Claude Code docs commands page is the authoritative list.

Command	What it does
/help	List all available commands
/init	Generate a starter CLAUDE.md for the project
/memory	Open and edit your memory and CLAUDE.md files
/compact	Compress the conversation to free up context
/clear	Reset the conversation for an unrelated task
/config	Open settings (model, theme, permissions)
/cost	Show token usage and cost for the session
/context	Show what files and context Claude is holding
/hooks	List active hooks
/agents	Manage and view subagents
/mcp	View connected MCP servers
/review	Review recent changes

Beyond the built-ins, you can write your own commands as markdown files in .claude/commands/<name>.md (or ~/.claude/commands/ for personal ones). The body is the prompt template and optional YAML frontmatter sets the description, allowed tools, and default model. Custom commands are where most of the productivity gain lives. Our Claude Code commands guide covers building your own.

The CLI flags cheatsheet

Slash commands run inside a session. CLI flags configure how claude starts, which is what you need for scripting and automation. These are the ones documented in the official CLI reference that you will reach for most.

Flag	Short	What it does
--print	-p	Print the response and exit (the basis of all scripting)
--continue	-c	Resume the most recent session in this directory
--resume	-r	Resume a specific past session
--model		Choose the model for the session (Opus, Sonnet, or Haiku)
--permission-mode		Set how Claude asks for permission to run tools
--add-dir		Add another directory to the working context
--output-format		Set output format (for example stream-json for pipelines)

For permission modes, default keeps the interactive prompts, acceptEdits auto-accepts file edits, and bypassPermissions skips prompts entirely. Use the last one only inside a sandboxed environment such as CI, never on a machine where Claude can touch anything you care about. Pair -p with --output-format stream-json when you want to feed Claude Code into your own harness.

The config files cheatsheet

Most of Claude Code lives in a .claude/ directory in your project, plus a ~/.claude/ directory for global settings. Knowing what each file does saves a lot of trips back to the docs.

Path	Purpose
CLAUDE.md	Project instructions Claude reads every session
~/.claude/CLAUDE.md	Global personal instructions across all projects
.claude/settings.json	Project settings, permissions, and hooks (team-shared)
.claude/settings.local.json	Personal overrides not committed to git
~/.claude/settings.json	Global user settings
.claude/commands/	Custom slash commands as markdown files
.claude/skills/	Reusable skill procedures
.claude/agents/	Subagent definitions
.mcp.json	MCP server configuration
.claudeignore	Files to exclude from Claude's reading (gitignore syntax)

A practical rule: anything you want shared with your team goes in .claude/ and gets committed; anything personal goes in settings.local.json or ~/.claude/. The setup guide walks through building each of these from scratch.

What the docs cost you nothing to learn, but the setup costs time

Reading the Claude Code docs is free. The real cost is the configuration work: a default install is a smart terminal assistant, but the autonomous, project-aware version takes weeks of writing CLAUDE.md files, building a skill library, wiring hooks, and tuning permissions. The docs tell you how each piece works; they cannot do the assembly for you.

That gap is where Claudify fits. It is a pre-built operating system for Claude Code: a large skill library, a memory architecture, specialist agents, slash commands, and quality-gate hooks, installed in one command. It is not a replacement for the official docs, it is the assembled configuration the docs describe, so you skip the weeks of setup and start from a working system.

Pricing context (so the docs make sense)

The docs assume you have access, which depends on your plan. Claude Code is included with Claude Pro and above, not the Free tier. Pro is $20 per month (or $17 per month billed annually). Max starts at $100 per month for 5x usage and goes up to around $200 per month for 20x. Team is roughly $25 per seat on the standard plan, or $125 per seat on the premium plan that includes Claude Code, and Enterprise is custom.

If you would rather pay per token through the API, that is billed separately: per million tokens, Opus is $15 input and $75 output, Sonnet is $3 and $15, and Haiku is $1 and $5. The current models are Opus 4.8, Sonnet 4.6, and Haiku 4.5. For the full breakdown, the official docs.claude.com pricing page is canonical, and our Claude Code pricing guide covers how the plans map to real usage.

Frequently asked questions

Where are the official Claude Code docs?

The official Claude Code docs are at docs.anthropic.com (also mirrored at docs.claude.com). Those pages cover the CLI reference and the commands list, which are the two sections most people return to once they are set up.

Is there a Claude Code desktop app, or is it only in the docs as a CLI?

Claude Code is a terminal CLI. There is no standalone Claude Code desktop app. It runs in your terminal and through IDE extensions for VS Code and JetBrains. The separate "Claude" desktop app is the chat product, not Claude Code, which is a common point of confusion in the docs.

What config files does Claude Code use?

The main ones are CLAUDE.md for instructions, .claude/settings.json for settings, permissions, and hooks, .claude/commands/ for custom commands, .claude/skills/ for skills, and .mcp.json for MCP servers. A global ~/.claude/ directory holds your personal defaults. The config files cheatsheet above lists each one.

Do I need to read all the docs to start?

No. Install Claude Code, run /init to create a CLAUDE.md, and start giving it tasks. Learn the agentic loop, CLAUDE.md, and a few slash commands first. Reach for the deeper documentation on hooks, MCP, and the SDK only when you need those specific features.

Is Claude Code free?

Claude Code is free to install, but it needs a paid plan to run because it uses Anthropic's models. It is included with Claude Pro and above, not the Free tier, or you can pay per token through the API. See our Claude Code pricing guide for the detail.

Keep this page bookmarked

The official Claude Code docs are the source of truth, but they are built for reference, not for a quick scan. Keep this page for the map and the cheatsheet: where each docs section lives, the core concepts that make the rest readable, and the commands, flags, and config files you reach for daily.

When you are ready to skip the weeks of manual configuration the docs describe, get Claudify: a complete Claude Code operating system, installed in one command.