Reference

CLI reference

Complete Claude Code CLI reference: all commands, flags, options, and usage examples.

Publié le 11 mars 2026

Main command: `claude`

claude [OPTIONS] [PROMPT]

General flags

Flag	Alias	Type	Description
`--model <id>`	`-m`	string	Model to use (`claude-haiku-4-5`, `claude-sonnet-4-5`, `claude-opus-4-5`, etc.)
`--print`	`-p`	boolean	Non-interactive mode: print the response and exit
`--output-format <fmt>`		string	Output format: `text` (default), `json`, `stream-json`
`--verbose`		boolean	Show tool calls and metadata (debug)
`--no-cache`		boolean	Disable prompt caching (avoid cached results)
`--dangerously-skip-permissions`		boolean	Not recommended: skip all confirmation prompts
`--continue`	`-c`	boolean	Resume the last conversation session
`--resume <id>`		string	Resume a specific session by its identifier
`--max-turns <n>`		number	Limit the number of exchange turns (agentic mode)
`--allowedTools <list>`		string	Allowed tools (comma-separated: `Bash,Read,Write`)
`--disallowedTools <list>`		string	Disallowed tools (comma-separated)
`--system-prompt <text>`	`-s`	string	System prompt to inject (overrides CLAUDE.md)
`--append-system-prompt <text>`		string	Text appended to the existing system prompt
`--add-dir <path>`		string	Add a directory to the accessible file context
`--debug`		boolean	Extended debug mode (internal logs)
`--version`	`-v`	boolean	Show the installed Claude Code version
`--help`	`-h`	boolean	Show help

--dangerously-skip-permissions

This flag bypasses all Claude Code confirmations, including before destructive actions (file deletion, script execution, API calls). Only use it in an isolated sandbox environment, never in production.

Examples: `claude` command

# Standard interactive session
claude

# Interactive session with a specific model
claude --model claude-opus-4-5

# Print mode: execute a task and exit
claude --print "Explain the code in src/auth.ts"
claude -p "Generate unit tests for utils/date.ts"

# Print mode with JSON format
claude --print --output-format json "List the functions in this file"

# Stream-json mode (for real-time processing)
claude --print --output-format stream-json "Analyze this project"

# Limit the number of turns (agentic mode)
claude --max-turns 5 "Refactor the auth module"

# Restrict available tools
claude --allowedTools "Read,Bash" "Analyze without modifying files"

# Resume the last session
claude --continue

# Resume a session by ID
claude --resume abc123

# Verbose (debug)
claude --verbose "Why is this test failing?"

Pipe mode (stdin -> stdout)

The --print mode is compatible with standard Unix pipes.

# Pass a file as input
cat src/auth.ts | claude --print "Add JSDoc comments to this code"

# Chain commands
git diff HEAD~1 | claude --print "Summarize these changes in bullet points"

# Pass a prompt via standard input
echo "Explain the Observer Design Pattern" | claude --print

# Integration in a shell script
RESULT=$(claude --print --output-format json "Analyze this code" < src/main.ts)
echo $RESULT | jq '.content[0].text'

# Read from a file
claude --print "What are the security risks here?" < config/database.yml

SDK / headless mode

For automated scripts, CI/CD pipelines, and applications that integrate Claude Code.

# Non-interactive execution with strict control
ANTHROPIC_API_KEY="sk-ant-..." \
  claude \
  --print \
  --max-turns 1 \
  --output-format json \
  "Check code quality in src/"

# Agentic mode with multiple turns
claude \
  --print \
  --max-turns 10 \
  --dangerously-skip-permissions \
  "Run the test suite and fix the errors"

# With environment variables in the config
claude \
  --print \
  --allowedTools "Read,Bash,Write" \
  --system-prompt "You are a security expert. Do not modify any files." \
  "Audit this project"

CI/CD integration

In CI/CD, always use --print and --max-turns to avoid infinite sessions. Set ANTHROPIC_API_KEY as a secret in your CI platform (GitHub Actions, GitLab CI, etc.). See our complete guide on hooks and headless mode.

Subcommand: `claude config`

Manages Claude Code configuration (~/.claude/settings.json file by default).

claude config [SUBCOMMAND] [OPTIONS]

Subcommand	Description
`claude config list`	List all active options and their values
`claude config get <key>`	Show the value of a specific key
`claude config set <key> <value>`	Set a key's value
`claude config add <key> <value>`	Add a value to an array
`claude config remove <key> <value>`	Remove a value from an array

# Examples
claude config list

claude config get model
# -> claude-sonnet-4-5

claude config set model claude-opus-4-5

claude config add allowedTools "Bash"
claude config remove allowedTools "Bash"

# With scope (global by default)
claude config --global set model claude-haiku-4-5

Subcommand: `claude mcp`

Manages MCP (Model Context Protocol) servers.

claude mcp [SUBCOMMAND] [OPTIONS]

`claude mcp add`

claude mcp add [OPTIONS] <name> -- <command> [args...]

Option	Description
`--scope <scope>`	`global` (default), `project`, or `user`
`--transport <type>`	`stdio` (default) or `sse`
`-e <VAR=value>`	Set an environment variable for the MCP

# Add a stdio MCP (local, via npx)
claude mcp add filesystem -- npx -y @modelcontextprotocol/server-filesystem ~/projects

# Add an MCP with project scope
claude mcp add --scope project postgres -- \
  npx -y @modelcontextprotocol/server-postgres postgresql://localhost/mydb

# Add an MCP with an environment variable
claude mcp add github -e GITHUB_TOKEN=ghp_xxx -- \
  npx -y @modelcontextprotocol/server-github

# Add an MCP via SSE transport (HTTP)
claude mcp add --transport sse my-api https://api.example.com/mcp

# Add a Python MCP (via uvx)
claude mcp add my-script -- uvx my-python-package

`claude mcp remove`

claude mcp remove [--scope <scope>] <name>

claude mcp remove filesystem
claude mcp remove --scope project postgres

`claude mcp list`

claude mcp list
# Shows: name, status (connected/disconnected), available tools, scope

`claude mcp logs`

# Show MCP logs (useful for debugging)
claude mcp logs github

# Follow logs in real time
claude mcp logs -f filesystem

`claude mcp get`

# Show the full configuration for an MCP
claude mcp get github

Subcommand: `claude doctor`

Diagnoses your Claude Code installation and configuration.

claude doctor

It checks:

Claude Code version and available updates
Network connectivity to the Anthropic API
API key validity (ANTHROPIC_API_KEY)
Status of configured MCPs (connection, available tools)
System permissions (file read/write)
Shell and environment variable configuration

# Example output
claude doctor
# -> Claude Code v1.x.x (up to date)
# -> Anthropic API reachable
# -> ANTHROPIC_API_KEY valid
# -> MCP filesystem: connected (5 tools)
# x MCP github: connection error (GITHUB_TOKEN missing)
# -> File permissions: OK

Subcommand: `claude update`

# Update Claude Code to the latest version
claude update

# Check version without updating
claude --version

Advanced options: output formats

`--output-format text` (default)

Plain text formatted for human reading. Includes markdown rendered in the terminal.

`--output-format json`

Returns a complete JSON object with the response and metadata.

{
  "type": "result",
  "subtype": "success",
  "cost_usd": 0.0023,
  "is_error": false,
  "duration_ms": 1842,
  "num_turns": 1,
  "result": "Response text here...",
  "session_id": "abc123"
}

`--output-format stream-json`

Stream of JSON messages (newline-delimited) in real time, suited for streaming processing.

# Line-by-line processing
claude --print --output-format stream-json "prompt" | while IFS= read -r line; do
  echo "$line" | jq -r 'select(.type == "content_block_delta") | .delta.text // empty'
done

Next steps

Cheatsheet: Quick reference: Slash commands and shortcuts
settings.json: Complete guide: All configuration options
Environment variables: Variables recognized by Claude Code

Claude Code cheatsheet

Settings reference

CLI reference

Main command: claude

General flags

Examples: claude command

Pipe mode (stdin -> stdout)

SDK / headless mode

Subcommand: claude config

Subcommand: claude mcp

claude mcp add

claude mcp remove

claude mcp list

claude mcp logs

claude mcp get

Subcommand: claude doctor

Subcommand: claude update