Skip to main content
The claude command starts an interactive session by default. Pass a prompt as an argument to send a single message, or use -p / --print for non-interactive pipeline usage. 
claude [prompt] [options]
claude <subcommand> [options]

​
Global flags

​
-p, --print
boolean
Print the response to stdout and exit immediately. Skips the workspace trust dialog. Only use this flag in directories you trust.
​
--model <model>
string
Model to use for the current session. Accepts a short alias (e.g. sonnet, opus) or a full model ID (e.g. claude-sonnet-4-6).
​
-d, --debug [filter]
string | boolean
Enable debug mode. Optionally pass a comma-separated category filter (e.g. "api,hooks" or "!1p,!file" to exclude categories).
​
--verbose
boolean
Override the verbose mode setting from config.
​
--permission-mode <mode>
string
Permission mode for the session. Choices: default, acceptEdits, bypassPermissions, dontAsk, plan.
​
--dangerously-skip-permissions
boolean
Bypass all permission checks. Recommended only for sandboxes with no internet access.
​
--settings <file-or-json>
string
Path to a settings JSON file, or a JSON string, to load additional settings from.
​
-r, --resume [session-id]
string | boolean
Resume a conversation. Provide a session ID to resume directly, or omit the value to open an interactive picker. Optionally pass a search term to pre-filter the picker.
​
-c, --continue
boolean
Continue the most recent conversation in the current directory.
​
--max-turns <turns>
number
Maximum number of agentic turns in non-interactive mode. Causes an early exit after the specified number of turns. Only works with --print.
​
--output-format <format>
string
Output format. Only works with --print. Choices:
  • text (default) — plain text
  • json — single JSON result
  • stream-json — realtime streaming JSON
​
--input-format <format>
string
Input format. Only works with --print. Choices:
  • text (default) — plain text
  • stream-json — realtime streaming JSON input
​
--add-dir <directories...>
string[]
Additional directories to allow tool access to.
​
--allowedTools, --allowed-tools <tools...>
string[]
Comma- or space-separated list of tool names to allow (e.g. "Bash(git:*) Edit").
​
--disallowedTools, --disallowed-tools <tools...>
string[]
Comma- or space-separated list of tool names to deny (e.g. "Bash(git:*) Edit").
​
--mcp-config <configs...>
string[]
Load MCP servers from JSON files or inline JSON strings (space-separated).
​
-n, --name <name>
string
Set a display name for this session (shown in /resume and the terminal title).
​
--plugin-dir <path>
string
Load plugins from a directory for this session only. Repeatable: --plugin-dir A --plugin-dir B.
​
--system-prompt <prompt>
string
System prompt to use for the session.
​
--append-system-prompt <prompt>
string
Append text to the default system prompt.
​
--no-session-persistence
boolean
Disable session persistence — sessions will not be saved to disk and cannot be resumed. Only works with --print.
​
--fallback-model <model>
string
Enable automatic fallback to the specified model when the default model is overloaded. Only works with --print.
​
--effort <level>
string
Effort level for the session. Choices: low, medium, high, max.
​
--bare
boolean
Minimal mode: skips hooks, LSP, plugin sync, attribution, auto-memory, background prefetches, keychain reads, and CLAUDE.md auto-discovery. Sets CLAUDE_CODE_SIMPLE=1. Auth is strictly ANTHROPIC_API_KEY or apiKeyHelper via --settings.

​
Usage examples

claude

​
Subcommands

SubcommandDescription
mcpConfigure and manage MCP servers
pluginManage Claude Code plugins
configOpen the interactive configuration panel
memoryEdit Claude memory files (CLAUDE.md)
sessionResume and manage past sessions
authManage authentication
doctorRun diagnostic checks
Subcommands for auth, doctor, and others follow the same help flag convention: use -h or --help to see their options.

Build docs developers (and LLMs) love

Get started for freeTalk to us