Claude Code CLI Flags Reference

A comprehensive, organized reference of command-line flags, subcommands, in-session slash commands, and environment variables for Claude Code, Anthropic's official CLI for Claude.

Sourced from the official documentation at code.claude.com/docs.

Note: Claude Code evolves quickly. Flags here reflect the docs as of the last update to this repo — always cross-check with claude --help or the official docs for the most current behavior.

Session Control
Output & Interactive Mode
Input Format
Model & Performance
Permissions & Security
System Prompt & Configuration
Working Directory & File Access
MCP (Model Context Protocol)
Agents & Subagents
Parallel & Background Execution
Initialization & Startup
Structured Output & Schema
Browser & IDE Integration
Remote & Web Sessions
Advisor Tool
Tmux & Multiplexing
Advanced Output & Streaming
Budget & Cost
Limits & Control
Debug & Version
Subcommands
In-Session Slash Commands
Environment Variables
Deprecated Flags & Aliases
Common Usage Patterns

Session Control

Flag	Short	Type	Description	Example
`--continue`	`-c`	flag	Load the most recent conversation and continue it	`claude -c`
`--resume`	`-r`	string (session ID/name)	Resume a specific session by ID or display name; omit value to show a picker	`claude -r "auth-refactor"`
`--fork-session`	—	flag	Create a new session ID when resuming (fork instead of continue)	`claude --resume abc123 --fork-session`
`--session-id`	—	UUID string	Use a specific session UUID instead of auto-generating one	`claude --session-id "550e8400-e29b-41d4-a716-446655440000"`
`--name`	`-n`	string	Set a human-readable display name for the session	`claude -n "my-feature-work"`
`--no-session-persistence`	—	flag	Don't save the session to disk (ephemeral)	`claude -p --no-session-persistence "query"`

Output & Interactive Mode

Flag	Short	Type	Description	Example
`--print`	`-p`	flag	Print the response and exit (non-interactive mode)	`claude -p "explain this function"`
`--output-format`	—	string	Output format: `text` (default), `json`, `stream-json`	`claude -p "query" --output-format json`
`--verbose`	—	flag	Enable verbose logging to stderr	`claude --verbose -p "query"`
`--ax-screen-reader`	—	flag	Render screen-reader-friendly output	`claude --ax-screen-reader`

Input Format

Flag	Type	Description	Example
`--input-format`	string	Input format: `text` (default), `stream-json`	`claude -p --input-format stream-json`
`--replay-user-messages`	flag	Re-emit user messages for acknowledgment in the stream	`claude -p --input-format stream-json --output-format stream-json --replay-user-messages`
`--include-partial-messages`	flag	Include partial streaming events in output	`claude -p --output-format stream-json --include-partial-messages "query"`
`--include-hook-events`	flag	Include all hook lifecycle events in the output stream	`claude -p --output-format stream-json --include-hook-events "query"`

Model & Performance

Flag	Type	Description	Example
`--model`	string	Set model: `sonnet`, `opus`, `haiku`, `fable`, or a full model ID	`claude --model claude-opus-5`
`--fallback-model`	string	Comma-separated fallback models, tried in order if the primary fails	`claude --fallback-model sonnet,haiku`
`--effort`	string	Reasoning effort level: `low`, `medium`, `high`, `xhigh`, `max`, `ultracode`	`claude --effort high`
`--betas`	string	Enable beta API headers for experimental features	`claude --betas interleaved-thinking`

Permissions & Security

Permission mode

Flag	Type	Description	Example
`--permission-mode`	string	Starting permission mode: `default` (prompt each time), `acceptEdits`, `plan`, `auto`, `dontAsk`, `bypassPermissions`, `manual`	`claude --permission-mode plan`
`--allow-dangerously-skip-permissions`	flag	Make `bypassPermissions` selectable without starting in it	`claude --permission-mode plan --allow-dangerously-skip-permissions`
`--dangerously-skip-permissions`	flag	Skip all permission prompts — disables safety gates. Use with caution, only in trusted/sandboxed environments.	`claude --dangerously-skip-permissions`
`--permission-prompt-tool`	string	Delegate permission prompts to an MCP tool	`claude -p --permission-prompt-tool mcp_auth_tool "query"`

Tool allow/deny lists

Flag	Type	Description	Example
`--allowedTools` / `--allowed-tools`	string array	Tools/patterns that run without prompting	`claude --allowed-tools "Bash(git log *)" "Read"`
`--disallowedTools` / `--disallowed-tools`	string array	Tools/patterns explicitly denied	`claude --disallowed-tools "Bash(rm *)" "Edit"`
`--tools`	string	Restrict available tools; empty = none, `default` = all, or comma-separated list	`claude --tools "Bash,Edit,Read"`

Safe mode

Flag	Type	Description	Example
`--safe-mode`	flag	Disable all customizations (hooks, skills, settings) for troubleshooting	`claude --safe-mode`

System Prompt & Configuration

Flag	Type	Description	Example
`--system-prompt`	string	Replace the entire default system prompt	`claude --system-prompt "You are a security expert"`
`--system-prompt-file`	path	Load system prompt from a file (replaces default)	`claude --system-prompt-file ./custom-prompt.txt`
`--append-system-prompt`	string	Append text to the default system prompt	`claude --append-system-prompt "Always cite file paths"`
`--append-system-prompt-file`	path	Append system prompt content from a file	`claude --append-system-prompt-file ./extra-rules.txt`
`--append-subagent-system-prompt`	string	Append text to every subagent's system prompt	`claude -p --append-subagent-system-prompt "Cite file paths" "query"`
`--exclude-dynamic-system-prompt-sections`	flag	Move machine-specific sections (cwd, env, tools) to the first message instead of the system prompt	`claude -p --exclude-dynamic-system-prompt-sections "query"`
`--settings`	path or JSON	Override settings for this session	`claude --settings ./settings.json`
`--setting-sources`	string	Comma-separated sources to load: `user`, `project`, `local`	`claude --setting-sources user,project`

Working Directory & File Access

Flag	Short	Type	Description	Example
`--add-dir`	—	path(s)	Add additional working directories for file access without moving cwd	`claude --add-dir ../apps ../lib`
`--worktree`	`-w`	string	Create or use a git worktree	`claude -w feature-auth`

MCP (Model Context Protocol)

Flag	Type	Description	Example
`--mcp-config`	path or JSON	Load MCP servers from a JSON configuration file	`claude --mcp-config ./mcp.json`
`--strict-mcp-config`	flag	Only use MCP servers specified in config; ignore defaults	`claude --strict-mcp-config --mcp-config ./mcp.json`
`--channels`	string	MCP servers to listen to for notifications	`claude --channels plugin:my-notifier@my-marketplace`
`--dangerously-load-development-channels`	string	Enable development/experimental MCP channels (requires confirmation)	`claude --dangerously-load-development-channels server:webhook`

Agents & Subagents

Flag	Type	Description	Example
`--agent`	string	Specify a custom agent for the session	`claude --agent my-custom-agent`
`--agents`	JSON	Define custom subagents dynamically	`claude --agents '{"reviewer":{"description":"Reviews code","prompt":"..."}}'`
`--teammate-mode`	string	Agent team display mode: `in-process`, `auto`, `tmux`, `iterm2`	`claude --teammate-mode auto`

Parallel & Background Execution

Flag	Type	Description	Example
`--bg` / `--background`	flag	Start as a background agent; return immediately	`claude --bg "investigate the flaky test"`
`--exec`	string	Run a shell command as a background job	`claude --bg --exec 'pytest -x'`
`--from-pr`	string	Filter sessions by GitHub PR (number or URL)	`claude --from-pr 123`

Initialization & Startup

Flag	Type	Description	Example
`--init`	flag	Run Setup hooks with the `init` matcher	`claude -p --init "query"`
`--init-only`	flag	Run Setup hooks and exit (don't start an interactive session)	`claude --init-only`
`--maintenance`	flag	Run Setup hooks with the `maintenance` matcher	`claude -p --maintenance "query"`
`--ide`	flag	Auto-connect to IDE on startup if available	`claude --ide`
`--bare`	flag	Minimal mode: skip auto-discovery of tools/extensions for faster startup	`claude --bare -p "query"`
`--disable-slash-commands`	flag	Disable all skills and `/` commands	`claude --disable-slash-commands`

Structured Output & Schema

Flag	Type	Description	Example
`--json-schema`	JSON	Request JSON output matching a JSON Schema	`claude -p --json-schema '{"type":"object","properties":{"name":{"type":"string"}}}'`

Browser & IDE Integration

Flag	Type	Description	Example
`--chrome`	flag	Enable Chrome browser integration for computer use	`claude --chrome`
`--no-chrome`	flag	Explicitly disable Chrome integration	`claude --no-chrome`
`--plugin-dir`	path	Load a custom plugin from a directory or `.zip` file	`claude --plugin-dir ./my-plugin`
`--plugin-url`	URL	Fetch a plugin from a remote URL	`claude --plugin-url https://example.com/plugin.zip`

Remote & Web Sessions

Flag	Type	Description	Example
`--cloud`	string	Create a web session on claude.ai and run the query there	`claude --cloud "Fix the login bug"`
`--remote`	string	Deprecated alias for `--cloud`	`claude --remote "Fix the login bug"`
`--teleport`	flag	Resume a web session (from claude.ai) locally	`claude --teleport`
`--remote-control` / `--rc`	string (optional)	Start with Remote Control enabled for claude.ai/code	`claude --remote-control "My Project"`
`--remote-control-session-name-prefix`	string	Prefix for auto-generated Remote Control session names	`claude remote-control --remote-control-session-name-prefix dev-box`

Advisor Tool

Flag	Type	Description	Example
`--advisor`	string	Enable the advisor tool with a specified model: `opus` or `sonnet`	`claude --advisor opus`

Tmux & Multiplexing

Flag	Type	Description	Example
`--tmux`	string (optional)	Create a tmux session; pass `classic` for traditional tmux behavior	`claude -w feature-auth --tmux`

Advanced Output & Streaming

Flag	Type	Description	Example
`--prompt-suggestions`	flag	Emit predicted next prompts in the output stream	`claude -p --prompt-suggestions --output-format stream-json --verbose "query"`
`--forward-subagent-text`	flag	Emit subagent text blocks in the output stream	`claude -p --output-format stream-json --forward-subagent-text "query"`

Budget & Cost

Flag	Type	Description	Example
`--max-budget-usd`	number	Set the maximum spend in USD for this session	`claude -p --max-budget-usd 5.00 "query"`

Limits & Control

Flag	Type	Description	Example
`--max-turns`	number	Limit agentic turns (number of model invocations)	`claude -p --max-turns 3 "query"`

Debug & Version

Flag	Short	Type	Description	Example
`--debug`	—	string (optional)	Enable debug logging; optionally filter by component	`claude --debug "api,mcp"`
`--debug-file`	—	path	Write debug logs to a specific file	`claude --debug-file /tmp/claude-debug.log`
`--version`	`-v`	flag	Print the Claude Code version and exit	`claude --version`

Subcommands

`claude mcp` — MCP server management

Subcommand	Purpose
`claude mcp reconnect`	Reconnect to all configured MCP servers
`claude mcp enable <server-name>`	Enable a specific MCP server
`claude mcp disable <server-name>`	Disable a specific MCP server

`claude config` — Settings management

Subcommand	Purpose
`claude config`	Open the settings editor
`claude config [key]`	Get the value of a specific key
`claude config [key=value]`	Set a specific key

`claude doctor` — Diagnostics & health check

Subcommand	Purpose
`claude doctor`	Run a health check on your Claude Code setup; diagnose & auto-fix issues

`claude update` — Version management

Subcommand	Purpose
`claude update`	Check for and install Claude Code updates
`claude update [version]`	Update to a specific version

`claude help` — Help & documentation

Subcommand	Purpose
`claude help`	Show all available commands & flags
`claude [command] --help`	Show help for a specific command

In-Session Slash Commands

Typed during an interactive session (not CLI flags, but included for a complete reference).

Session control

Command	Args	Purpose
`/clear`	`[name]`	Start a new conversation; optionally name the old one
`/resume`	—	Show a picker to resume an earlier conversation
`/branch`	`[name]`	Create a new conversation branch from the current point
`/fork`	`[prompt]`	Copy the conversation to a new background session
`/cd`	`<path>`	Move the session to a new working directory
`/add-dir`	`<path>`	Add a directory for file access without moving cwd
`/exit`	—	Exit the CLI (keeps background sessions running)

Code review & analysis

Command	Args & flags	Purpose
`/code-review`	`[level] [--fix] [--comment] [target]`	Review a diff for bugs & cleanups; levels: `low`, `medium`, `high`, `xhigh`, `max`, `ultra`
`/simplify`	—	Cleanup-only code review (no bug hunting)
`/security-review`	—	Security vulnerability check on the diff
`/diff`	—	Interactive diff viewer for uncommitted changes

Model & performance control

Command	Args	Purpose
`/model`	`[model-name]`	Switch AI model (e.g. `opus`, `sonnet`, `haiku`)
`/effort`	`[level\|auto]`	Set reasoning effort: `low`, `medium`, `high`, `xhigh`, `max`, `ultracode`
`/advisor`	`[model\|off]`	Enable/disable the advisor tool
`/fast`	`[on\|off]`	Toggle fast mode for quicker responses

Context & memory

Command	Args	Purpose
`/context`	`[all]`	Visualize context usage as a colored grid
`/compact`	`[instructions]`	Summarize the conversation to free context space
`/btw`	`<question>`	Ask a quick side question without adding it to history
`/memory`	—	Edit memory files, manage auto-memory

Parallel & background work

Command	Args	Purpose
`/batch`	`<instruction>`	Decompose a large change into parallel units
`/background`	`[prompt]`	Detach the session as a background agent
`/tasks`	—	List the current session's background work
`/subtask`	—	Hand a side task to a subagent

Documentation & reference

Command	Args	Purpose
`/init`	—	Initialize the project with a `CLAUDE.md` guide
`/claude-api`	`[migrate\|managed-agents-onboard]`	Load Claude API reference & docs
`/plan`	—	Switch to plan mode for large changes

Workflow & automation

Command	Args	Purpose
`/loop`	`[interval] [prompt]`	Run a prompt repeatedly on a schedule (default 10m)
`/goal`	`[condition\|clear]`	Set a condition for autonomous work
`/deep-research`	`<question>`	Fan out web searches with a cited report

Development & debugging

Command	Args	Purpose
`/debug`	`[description]`	Enable debug logging & troubleshoot
`/doctor`	—	Run a setup checkup; diagnose & fix issues
`/mcp`	`[reconnect\|enable\|disable]`	Manage MCP server connections
`/hooks`	—	View hook configurations

Utilities

Command	Args	Purpose
`/copy`	`[N]`	Copy the last response (or Nth-latest) to clipboard
`/export`	`[filename]`	Export the conversation as plain text
`/config`	`[key=value]`	Open settings or set a specific key
`/color`	`[color\|default]`	Set the prompt bar color
`/permissions`	—	Set approval rules

Integration

Command	Purpose
`/install-github-app`	Install the Claude GitHub App
`/install-slack-app`	Install the Claude Slack app (web sessions only)
`/design-sync`	Upload a React design system to Claude Design
`/autofix-pr`	Watch a PR & push fixes when CI fails

Account & help

Command	Purpose
`/login`	Sign in to your Anthropic account
`/logout`	Sign out
`/help`	Show available commands
`/feedback`	Send product feedback
`/bug`	Report a bug with session context

Environment Variables

Set in your shell or in ~/.claude/settings.json under the env key.

Authentication & API

Variable	Type	Description
`ANTHROPIC_API_KEY`	string	API key for Anthropic authentication
`ANTHROPIC_AUTH_TOKEN`	string	Custom Authorization header value
`ANTHROPIC_BASE_URL`	URL	Override the API endpoint (for proxies)
`ANTHROPIC_MODEL`	string	Default model to use

AWS & cloud integration

Variable	Type	Description
`ANTHROPIC_AWS_API_KEY`	string	Workspace API key for Claude Platform on AWS
`ANTHROPIC_AWS_BASE_URL`	URL	Override the AWS endpoint
`ANTHROPIC_AWS_WORKSPACE_ID`	string	Workspace ID for AWS (required for AWS integration)
`ANTHROPIC_BEDROCK_BASE_URL`	URL	Override the Amazon Bedrock endpoint
`ANTHROPIC_VERTEX_BASE_URL`	URL	Override the Google Cloud Agent Platform endpoint

Timeout & performance

Variable	Type	Default	Description
`API_TIMEOUT_MS`	number	600000 (10 min)	Request timeout in milliseconds
`API_FORCE_IDLE_TIMEOUT`	number	—	Override the 5-minute idle timeout for streaming
`BASH_DEFAULT_TIMEOUT_MS`	number	120000 (2 min)	Default timeout for bash commands
`BASH_MAX_TIMEOUT_MS`	number	600000 (10 min)	Maximum timeout the model can set for bash

Feature toggles

Variable	Type	Description
`CLAUDE_CODE_DISABLE_ARTIFACT`	flag	Disable artifact publishing
`CLAUDE_CODE_DISABLE_AGENT_VIEW`	flag	Disable the background agents display
`CLAUDE_CODE_DISABLE_AUTO_MEMORY`	flag	Disable auto-memory functionality
`CLAUDE_CODE_DISABLE_BACKGROUND_TASKS`	flag	Disable background task functionality
`CLAUDE_CODE_AUTO_CONNECT_IDE`	flag	Auto-connect to IDE (override detection)
`CLAUDE_CODE_DISABLE_ALTERNATE_SCREEN`	flag	Use the classic renderer instead of fullscreen
`CLAUDE_CODE_DISABLE_ATTACHMENTS`	flag	Disable file attachment processing

Debugging & logging

Variable	Type	Description
`CLAUDE_CODE_DEBUG_LOGS_DIR`	path	Override the debug log file directory
`CLAUDE_CODE_DEBUG_LOG_LEVEL`	string	Minimum log level: `verbose`, `debug`, `info`, `warn`, `error`
`DEBUG`	string	Enable debug mode; use with `--debug` flag

Setting variables

Shell (macOS/Linux/WSL):

export API_TIMEOUT_MS="1200000"
export ANTHROPIC_MODEL="claude-opus-5"
claude

Shell (Windows PowerShell):

$env:API_TIMEOUT_MS = "1200000"
$env:ANTHROPIC_MODEL = "claude-opus-5"
claude

Settings file (~/.claude/settings.json):

{
  "env": {
    "API_TIMEOUT_MS": "1200000",
    "BASH_DEFAULT_TIMEOUT_MS": "300000",
    "ANTHROPIC_MODEL": "claude-opus-5"
  }
}

Precedence order (highest to lowest):

CLI flags
Environment variables (shell)
Settings file env key
Managed settings (organization-level)
Defaults

Deprecated Flags & Aliases

Deprecated	Replacement	Notes
`--remote`	`--cloud`	Renamed for clarity

Common Usage Patterns

# Basic interactive session
claude

# Non-interactive query with output
claude -p "explain this function"

# Continue last session
claude -c

# Resume named session
claude -r "session-name"

# Resume and fork (new session ID)
claude -r "session-name" --fork-session

# Set permissions and model
claude --permission-mode plan --model opus

# Custom system prompt (non-interactive)
claude --system-prompt "You are a security expert" -p "audit this code"

# JSON output
claude -p --output-format json "query"

# Structured output with schema
claude -p --json-schema '{"type":"object","properties":{"code":{"type":"string"}}}' "write hello world"

# Background execution (returns immediately)
claude --bg "run all tests" --effort high

# Debug mode with filtering
claude --debug "api,mcp" -p "query"

# Create a git worktree
claude -w feature-auth

# Add additional directories
claude --add-dir ../apps ../lib -p "analyze the codebase"

# Limit agentic turns
claude -p --max-turns 3 "solve this problem"

# Disable all customizations
claude --safe-mode

# Use a web session
claude --cloud "Fix the login bug"

# Chrome integration for computer use
claude --chrome -p "take a screenshot"

Contributing

Found a flag that's missing, wrong, or changed? Open a PR or issue. Please verify against claude --help or the official docs before submitting.

Name		Name	Last commit message	Last commit date
Latest commit History 4 Commits
system-prompts		system-prompts
PLUGIN-CATALOG.md		PLUGIN-CATALOG.md
README.md		README.md

Folders and files

Latest commit

History

Repository files navigation