Getting started

Introduction

Claude Code is Anthropic's AI coding assistant for the terminal. This reference covers CLI commands, flags, interactive mode, configuration, and workflows.

Install

curl -fsSL https://claude.ai/install.sh | bash  # native
brew install --cask claude-code                 # Homebrew
npm install -g @anthropic-ai/claude-code        # npm, Node 22+
claude install stable                           # stable channel

On Windows: irm https://claude.ai/install.ps1 | iex in PowerShell, or winget install Anthropic.ClaudeCode.

See: Setup

Sign in

claude                       # first run opens a browser
claude auth login            # sign in again
claude auth login --console  # bill to a Console account
claude auth status           # JSON status; --text for plain

See: Authentication

Sessions

claude                        # start interactive session
claude "explain this repo"    # start with a prompt
claude -p "summarize README"  # print mode: answer and exit
git diff | claude -p "review this diff"

Use claude -c to continue the last conversation, or claude -r "auth-refactor" to resume one by name. See: Quickstart

CLI reference

Commands

Sessions

CommandDescription
claudeStart interactive session
claude "query"Start with an initial prompt
claude -p "query"Print response and exit
claude -cContinue most recent conversation
claude -r <session>Resume a session by ID or name
claude --bg "task"Start a background agent

Setup

CommandDescription
claude updateUpdate to latest version
claude installInstall or pin a version
claude doctorCheck install and settings
claude importImport another agent's setup
claude mcpConfigure MCP servers
claude pluginInstall and manage plugins
claude auth statusShow authentication status
claude setup-tokenGenerate a token for CI

See: CLI reference

Flags

FlagDescription
-c, --continueContinue most recent conversation
-r, --resumeResume a session by ID or name
--model sonnetChoose the model
--effort highSet reasoning effort
--fallback-model sonnetFall back when overloaded
--permission-mode planStart in a permission mode
--add-dir ../libAdd extra working directories
--verboseShow full turn-by-turn output

See: CLI reference

Other flags

FlagDescription
--debugEnable debug mode
--settings file.jsonLoad settings for this session
--ideAuto-connect to your IDE
-w, --worktreeStart in an isolated git worktree
--dangerously-skip-permissionsSkip all permission prompts
--safe-modeStart with customizations off

See: CLI reference

Headless mode

FlagDescription
-p, --printPrint the response and exit
--output-format jsontext, json, or stream-json
--json-schema schemaConstrain output to a schema
--max-turns 3Limit agentic turns
--max-budget-usd 5Stop the run at a spend cap
--bareSkip hooks, plugins, and memory

See: Headless mode

Slash commands

Everyday commands

CommandDescription
/helpShow available commands
/clearStart a new conversation
/compactSummarize to free up context
/contextShow context window usage
/planEnter plan mode
/modelSwitch the AI model
/permissionsAllow, ask, and deny rules
/statusShow version, model, account

See: Commands

Project commands

CommandDescription
/initGenerate a starter CLAUDE.md
/memoryEdit memory files
/mcpManage MCP connections
/hooksView hook configurations
/doctorRun a setup checkup
/diffReview working-tree changes
/code-reviewReview the diff or a PR

See: Commands

Interactive mode

Keyboard shortcuts

Session

ShortcutDescription
Ctrl+CInterrupt, clear input, or exit
Ctrl+DExit Claude Code
EscInterrupt Claude, or close dialogs
Esc EscClear the draft, or rewind
Ctrl+BBackground running tasks
Ctrl+X Ctrl+KStop all background subagents; press twice

Prompt

ShortcutDescription
Shift+TabCycle permission modes
Alt+PSwitch model
Alt+TToggle extended thinking
Alt+OToggle fast mode
Ctrl+VPaste an image from the clipboard
Ctrl+GEdit the prompt in $EDITOR
Ctrl+SStash or restore the prompt

Display

ShortcutDescription
Ctrl+LRedraw the screen
Ctrl+RSearch prompt history
Ctrl+OToggle the transcript viewer
Ctrl+TToggle Claude's task list

On macOS, use Option instead of Alt; some chords need Option mapped to Meta.

See: Interactive mode

Input modes

InputEffect
/Slash commands and skills
!Run a shell command directly
@Mention a file path
:Insert an emoji by shortcode
?Show the shortcut help panel

For a newline, \ + Enter or Ctrl+J works anywhere; Shift+Enter needs /terminal-setup in some terminals.

See: Interactive mode

Configuration

Settings files

~/.claude/settings.json                # user settings
.claude/settings.json                  # project settings, committed
.claude/settings.local.json            # personal project settings
/etc/claude-code/managed-settings.json # managed policy (Linux)

Run /config to edit settings, or pass --settings file.json for one session. See: Settings

Environment variables

VariablePurpose
ANTHROPIC_API_KEYAPI key used instead of a subscription
ANTHROPIC_MODELDefault model name
ANTHROPIC_BASE_URLRoute requests through a gateway
API_TIMEOUT_MSAPI timeout (default 600000)
BASH_DEFAULT_TIMEOUT_MSBash tool timeout (default 120000)
CLAUDE_CODE_DISABLE_AUTO_MEMORYTurn auto memory off
DISABLE_AUTOUPDATERTurn background updates off

Settings files can also set env vars under an env key. See: Environment variables

Memory

CLAUDE.md             # project instructions, committed
CLAUDE.local.md       # personal notes, not committed
.claude/rules/*.md    # topic rules, optionally path-scoped
~/.claude/CLAUDE.md   # instructions for all your projects

/init generates a project CLAUDE.md; /memory edits memory files and auto memory. Import files with @path/to/file.md. See: Memory

Permission modes

ModeWhat runs without asking
defaultReads only (labeled Manual)
acceptEditsReads, file edits, filesystem commands
planReads, until you approve a plan
autoEverything, with safety checks
dontAskPre-approved tools only; rest denied
bypassPermissionsEverything

auto is the built-in starting mode; Shift+Tab cycles modes in a session. A classifier can still block actions, and explicit ask rules still prompt. See: Permission modes

Common workflows

Reviewing code

git diff | claude -p "review this diff for bugs"
claude -p "check src/ for security issues"

In a session, /code-review reviews the diff, /code-review 1234 a pull request, and /security-review scans for vulnerabilities.

See: Review locally with /code-review

Scripting

# One-shot JSON output for scripts
claude -p "list the failing tests" --output-format json
# Pipe logs in
cat build.log | claude -p "explain this failure"

--bare skips hooks, plugins, and auto-discovered config so scripted calls start faster. It ignores OAuth credentials, so set ANTHROPIC_API_KEY.

See: Headless mode

Background and parallel work

claude --bg "migrate the auth tests"  # detach immediately
claude agents                         # monitor and dispatch

Inside a session, /background detaches it and /batch fans work out across worktrees.

See: Agent view

Also see

0 Comments for this cheatsheet. Write yours!