CLI Flags

Run claude [prompt] to start an interactive session, or combine flags to control behavior. Use -p / --print for non-interactive (pipe-friendly) output.

claude [prompt] [flags]

Flags marked print-only only take effect when --print (-p) is also passed. Using them in interactive mode has no effect.

General

-h, --help

Type: booleanDisplay help for the command and exit.

claude --help

-p, --print

Type: booleanPrint the response to stdout and exit instead of starting an interactive session. Useful for pipes and scripts.

claude -p "Summarize this file" < README.md

The workspace trust dialog is skipped in --print mode. Only use this flag in directories you trust.

--verbose

Type: booleanOverride the verbose mode setting from config. Enables extra output including tool inputs and outputs.

claude --verbose -p "List files in src/"

-d, --debug [filter]

Type: boolean or stringEnable debug logging. Pass an optional comma-separated filter to restrict which categories are logged.

# All debug output
claude --debug

# Only API and hooks categories
claude --debug "api,hooks"

# Everything except 1p and file categories
claude --debug "!1p,!file"

--debug-file <path>

Type: stringWrite debug logs to a file instead of (or in addition to) the terminal. Implicitly enables debug mode.

claude --debug-file /tmp/claude-debug.log

--bare

Type: booleanMinimal startup mode. Disables hooks, LSP, plugin sync, commit attribution, auto-memory, background prefetches, keychain reads, and CLAUDE.md auto-discovery. Sets CLAUDE_CODE_SIMPLE=1.In --bare mode:

Anthropic auth is strictly ANTHROPIC_API_KEY or apiKeyHelper via --settings. OAuth and keychain are never read.
Third-party providers (Bedrock, Vertex, Foundry) use their own credentials as normal.
Skills still resolve via /skill-name.

Provide context explicitly using: --system-prompt, --append-system-prompt, --add-dir, --mcp-config, --settings, --agents, --plugin-dir.

claude --bare --settings settings.json -p "Review this code"

Session control

-c, --continue

Type: booleanContinue the most recent conversation in the current working directory.

claude -c
claude -c -p "Keep going"

-r, --resume [value]

Type: boolean or stringResume a conversation. Pass a session UUID to jump directly to that session, a search term to pre-filter the picker, or omit the value to open the interactive picker.

# Open picker
claude --resume

# Filter picker
claude -r "my-feature"

# Jump directly to session
claude -r 550e8400-e29b-41d4-a716-446655440000

--fork-session

Type: booleanWhen used with --resume or --continue, creates a new forked session from that point instead of continuing the original. Combine with --session-id to specify the ID of the new fork.

claude --resume --fork-session

--from-pr [value]

Type: boolean or stringResume a session linked to a GitHub PR. Pass a PR number or URL to jump directly, or omit to open the interactive picker.

claude --from-pr 123
claude --from-pr https://github.com/owner/repo/pull/123

--session-id <uuid>

Type: stringAssign a specific UUID to the new session. Must be a valid UUID and must not already be in use.

claude --session-id 550e8400-e29b-41d4-a716-446655440000

When used with --resume or --continue, requires --fork-session to be present.

-n, --name <name>

Type: stringSet a display name for the session. Shown in /resume and in the terminal title bar.

claude -n "refactor-auth-module"

--no-session-persistence

Type: boolean (print-only)Disable session persistence. The session will not be saved to disk and cannot be resumed.

claude -p --no-session-persistence "One-shot task"

Model and compute

--model <model>

Type: stringSet the model for the session. Accepts an alias for the latest model version (sonnet, opus) or a full model name (claude-sonnet-4-6).

claude --model sonnet
claude --model claude-opus-4-6

--effort <level>

Type: string — one of low, medium, high, maxSet the reasoning effort level for the session.

claude --effort high -p "Analyze this architecture"

--agent <agent>

Type: stringOverride the agent setting for this session. Selects which agent definition to use.

claude --agent reviewer

--agents <json>

Type: string (JSON)Inline agent definitions as a JSON object. Each key is the agent name; the value is an object with description and prompt.

claude --agents '{"reviewer": {"description": "Reviews code", "prompt": "You are a code reviewer"}}'

--fallback-model <model>

Type: string (print-only)Enable automatic fallback to the specified model when the primary model is overloaded.

claude --model opus --fallback-model sonnet -p "Analyze this"

--betas <betas...>

Type: string list (API key users only)Pass beta feature headers to include in API requests.

claude --betas interleaved-thinking-2025-05-14

Output format

--output-format <format>

Type: string — one of text, json, stream-json (print-only)Control the output format when running in --print mode.

Value	Description
`text`	Plain text (default)
`json`	Single JSON object with the complete result
`stream-json`	Newline-delimited JSON events streamed in real time

claude -p --output-format json "List the top 3 issues"
claude -p --output-format stream-json "Implement the feature"

--input-format <format>

Type: string — one of text, stream-json (print-only)Control how stdin is interpreted when running in --print mode.

Value	Description
`text`	Plain text (default)
`stream-json`	Newline-delimited JSON event stream

cat events.ndjson | claude -p --input-format stream-json --output-format stream-json ""

--include-hook-events

Type: boolean (requires --output-format=stream-json)Include all hook lifecycle events in the output stream. Without this flag only SessionStart and Setup events are emitted.

claude -p --output-format stream-json --include-hook-events "Run tests"

--json-schema <schema>

Type: string (JSON Schema)Provide a JSON Schema for structured output validation. Claude will produce output that conforms to the schema.

claude -p --json-schema '{"type":"object","properties":{"name":{"type":"string"}},"required":["name"]}' \
  "Extract the author name from this file"

--max-turns <n>

Type: integer (print-only)Limit the number of agentic turns before stopping. Useful to cap costs in automated pipelines.

claude -p --max-turns 5 "Fix the failing tests"

--max-budget-usd <amount>

Type: number — must be greater than 0 (print-only)Set a hard budget cap in USD. Claude stops before exceeding this amount.

claude -p --max-budget-usd 0.50 "Refactor this module"

--replay-user-messages

Type: boolean (requires --input-format=stream-json and --output-format=stream-json)Re-emit user messages from stdin back on stdout for acknowledgment.

claude -p --input-format stream-json --output-format stream-json --replay-user-messages ""

Tools and permissions

--allowed-tools <tools...>

Aliases: --allowedToolsType: string listComma or space-separated list of tool names to allow. Supports glob-style patterns.

claude --allowed-tools "Bash(git:*) Edit Read"
claude --allowed-tools Bash Edit

--disallowed-tools <tools...>

Aliases: --disallowedToolsType: string listComma or space-separated list of tool names to deny.

claude --disallowed-tools "Bash(rm:*)"

--tools <tools...>

Type: string listSpecify the exact set of built-in tools to make available. Use "" to disable all tools, "default" for all tools, or list tool names explicitly.

# No tools
claude --tools ""

# All built-in tools
claude --tools default

# Specific tools only
claude --tools "Bash,Edit,Read"

--permission-mode <mode>

Type: stringSet the permission mode for the session. Controls how Claude handles tool approval prompts.

claude --permission-mode auto

Run claude --help to see the current list of valid permission mode values.

--dangerously-skip-permissions

Type: booleanSkip all permission prompts. All tool calls are approved automatically.

Only use this in fully sandboxed or trusted automated environments. This flag removes all safety guardrails.

claude -p --dangerously-skip-permissions "Run the full test suite"

MCP configuration

--mcp-config <configs...>

Type: string listLoad MCP servers from one or more JSON configuration files or inline JSON strings (space-separated).

# From a file
claude --mcp-config ./mcp-servers.json

# Multiple files
claude --mcp-config ./servers-a.json ./servers-b.json

--strict-mcp-config

Type: booleanOnly use MCP servers defined via --mcp-config. Ignores all project-level and user-level MCP configurations.

claude --strict-mcp-config --mcp-config ./servers.json

System prompt

--system-prompt <prompt>

Type: stringReplace the default system prompt with a custom prompt for the session.

claude --system-prompt "You are a security auditor. Focus only on vulnerabilities."

--append-system-prompt <prompt>

Type: stringAppend additional text to the default system prompt without replacing it.

claude --append-system-prompt "Always output results as a numbered list."

File and directory access

--add-dir <directories...>

Type: string listGrant tool access to additional directories beyond the current working directory.

claude --add-dir /shared/libs --add-dir /shared/configs

--file <specs...>

Type: string listDownload file resources at startup. Format: file_id:relative_path.

claude --file file_abc123:doc.txt file_def456:image.png

Settings

--settings <file-or-json>

Type: stringLoad additional settings from a JSON file path or an inline JSON string.

# From a file
claude --settings ./my-settings.json

# Inline JSON
claude --settings '{"model": "sonnet", "verbose": true}'

--setting-sources <sources>

Type: stringComma-separated list of setting sources to load. Accepted values: user, project, local.

# Load only user-level settings (ignore project and local)
claude --setting-sources user

# Load user and project settings
claude --setting-sources user,project

Plugins

--plugin-dir <path>

Type: string (repeatable)Load plugins from a directory for this session only. Repeat the flag to specify multiple directories.

claude --plugin-dir ./my-plugins
claude --plugin-dir ./plugins-a --plugin-dir ./plugins-b

--disable-slash-commands

Type: booleanDisable all skills (slash commands loaded from skills/ directories and plugins).

claude --disable-slash-commands -p "Do this without any custom skills"

IDE integration

--ide

Type: booleanAutomatically connect to an IDE on startup if exactly one valid IDE integration is available.

claude --ide

Chrome integration

--chrome

Type: booleanEnable the Claude in Chrome integration.

claude --chrome

--no-chrome

Type: booleanDisable the Claude in Chrome integration.

claude --no-chrome

SDK

CLI Reference

General

Session control

Model and compute

Output format

Tools and permissions

MCP configuration

System prompt

File and directory access

Settings

Plugins

IDE integration

Chrome integration

Build docs developers (and LLMs) love

SDK

CLI Reference

​General

​Session control

​Model and compute

​Output format

​Tools and permissions

​MCP configuration

​System prompt

​File and directory access

​Settings

​Plugins

​IDE integration

​Chrome integration

Build docs developers (and LLMs) love

General

Session control

Model and compute

Output format

Tools and permissions

MCP configuration

System prompt

File and directory access

Settings

Plugins

IDE integration

Chrome integration