📑 Table of Contents
- Quick Start Commands
- CLI Flags (Launch-Time)
- Environment Variables
- Authentication & API
- Cloud Providers
- Model Configuration
- Agent Teams (Experimental)
- Bash & Shell Behavior
- Memory & Context
- MCP Protocol
- Tasks & Scheduling
- Plugins & Updates
- Disabling Features
- Network & Proxy
- Miscellaneous
- Slash Commands
- Hook Events
- Variable Substitutions
- Settings.json Reference
- Quick Reference Matrix
- Installation & Setup
- Tips & Best Practices
- Real-World Examples
Quick Start Commands
🎯 Most Used
| Command | Description | Example |
|---|---|---|
| claude | Start interactive REPL session | claude |
| claude "query" | Start session with initial prompt | claude "explain this function" |
| claude -p "query" | Non-interactive print mode (query and exit) | claude -p "review this code" |
| claude -c | Continue most recent conversation | claude -c |
| claude --continue | Long form of -c | claude --continue |
| claude -r "id" "query" | Resume a specific session by ID | claude -r "abc123" "finish the task" |
| claude --resume | Resume last session interactively | claude --resume |
| claude -v | Show installed version number | claude -v |
| cat file | claude -p "q" | Process piped stdin | cat error.log | claude -p "diagnose" |
| claude update | Update to latest version | claude update |
| claude install --force | Force reinstall (bypasses stale locks) | claude install --force |
| claude auth login | Authenticate with Anthropic account | claude auth login |
| claude auth login --console | Auth via API Console / billing account | claude auth login --console |
| claude auth status | Check current authentication status | claude auth status |
$ claude -p "list all TODO comments in this project"
$ cat error.log | claude -p "what caused this?"
$ claude --continue "complete the implementation"
📟 CLI Subcommands
| Command | Description | Example |
|---|---|---|
| claude auth login | Authenticate via claude.ai subscription | claude auth login |
| claude auth login --console | Authenticate via API Console/billing account | claude auth login --console |
| claude auth status | Check current authentication status | claude auth status |
| claude auth logout | Sign out of current account | claude auth logout |
| claude update | Update to latest version | claude update |
| claude install --force | Force reinstall, bypasses stale lock files | claude install --force |
| claude agents NEW | List all configured agents for the current project | claude agents |
| claude remote-control NEW | Start the remote control server as a standalone daemon | claude remote-control |
| claude mcp NEW | MCP management subcommands: add, remove, list, enable, disable, auth | claude mcp list |
| claude mcp add <name> <cmd> | Add a new MCP server to config (stdio transport) | claude mcp add myserver node server.js |
| claude mcp add --transport sse <url> | Add an SSE-based MCP server | claude mcp add --transport sse https://mcp.example.com |
| claude mcp remove <name> | Remove a named MCP server from config | claude mcp remove myserver |
| claude mcp auth <name> | Manually trigger OAuth flow for a named MCP server | claude mcp auth myserver |
CLI Flags (Launch-Time)
⚙️ Configuration
Passed when launching Claude Code to customize session behavior.
📂 Working Directory & Scope
| Flag | Description | Example |
|---|---|---|
| --add-dir | Add additional working directories (validates paths exist) | claude --add-dir ../apps ../lib |
| -w, --worktree | Launch in isolated git worktree (loads skills/hooks from worktree dir) | claude --worktree |
| --fork-session | Create new session ID when resuming (parallel work branches) | claude --fork-session |
🤖 Model & Agents
| Flag | Description | Example |
|---|---|---|
| --model | Set model (aliases: sonnet, opus, haiku — or full model ID) | claude --model claude-opus-4-6 |
| --agent | Specify a named agent for the session | claude --agent general-purpose |
| --teammate-mode | Set team display: auto | in-process | tmux | claude --teammate-mode in-process |
| --agents | Define custom subagents inline via JSON array | claude --agents '[{"name":"explorer"}]' |
| --effort | Set thinking effort: low | medium | high | max | auto | claude --effort high |
🔐 Permissions & Security
| Flag | Description | Example |
|---|---|---|
| --permission-mode | Start in a specific mode (e.g., plan) | claude --permission-mode plan |
| --enable-auto-mode NEW 2.1.82 | Enable auto mode (research preview) — Claude auto-approves low-risk actions via Sonnet 4.6 classifier. Middle ground between default and --dangerously-skip-permissions. Disable organisation-wide with managed setting disableAutoMode | claude --enable-auto-mode |
| --dangerously-skip-permissions | Skip all permission prompts ⚠️ — use only in devcontainer with firewall | claude --dangerously-skip-permissions |
| --allow-dangerously-skip-permissions | Allow the skip-permissions option without activating it | claude --permission-mode plan --allow-dangerously-skip-permissions |
| --tools | Restrict built-in tools (works in interactive AND -p mode) | claude --tools "Bash,Edit,Read" |
| --allowedTools | Tools that run without prompting for permission | claude --allowedTools "Write" "Bash(git *)" |
| --disallowedTools | Tools removed from context entirely; cannot be used | claude --disallowedTools "Bash(rm *)" |
| --permission-prompt-tool | MCP tool to handle permission prompts in -p (non-interactive) mode | claude -p --permission-prompt-tool mcp_auth_tool "q" |
💾 System Prompt Customization
| Flag | Description | Example |
|---|---|---|
| --system-prompt | Replace entire system prompt with custom text | claude --system-prompt "You are a Python expert" |
| --system-prompt-file | Load system prompt replacement from file (mutually exclusive with --system-prompt) | claude --system-prompt-file ./prompts/custom.txt |
| --append-system-prompt | Append text to the default system prompt (preserves built-in capabilities) | claude --append-system-prompt "Always use TypeScript" |
| --append-system-prompt-file | Append file contents to the default system prompt | claude --append-system-prompt-file ./style-rules.txt |
💡 Tip: All four system prompt flags work in both interactive and -p modes. Append flags preserve built-in capabilities — prefer them over replacement flags unless you need full control.
📤 Output & Input Formats
| Flag | Description | Example |
|---|---|---|
| --output-format | text | json | stream-json | claude -p --output-format json "query" |
| --input-format | Set input format (e.g., stream-json for piped input) | claude -p --input-format stream-json |
| --include-partial-messages | Include partial streaming events in output | claude -p --output-format stream-json --include-partial-messages "q" |
| --json-schema | Return validated JSON matching a specific schema | claude -p --json-schema '{"type":"object"}' "query" |
🔌 MCP & Plugins
| Flag | Description | Example |
|---|---|---|
| --mcp-config | Load MCP server definitions from a JSON file | claude --mcp-config ./mcp.json |
| --strict-mcp-config | Only use servers from --mcp-config; ignore all others | claude --strict-mcp-config --mcp-config ./mcp.json |
| --plugin-dir | Load plugins from directory for this session (flag is repeatable) | claude --plugin-dir ./plugins --plugin-dir ./more |
🔗 Web, Remote & Browser
| Flag | Description | Example |
|---|---|---|
| --remote | Create a new web session on claude.ai | claude --remote |
| --remote-control | Bridge terminal session to claude.ai for browser/phone control | claude --remote-control |
| --rc | Alias for --remote-control | claude --rc |
| --teleport | Resume a web/iOS running session in your local terminal | claude --teleport |
| --chrome | Enable Claude in Chrome browser integration (claude.ai/chrome extension) | claude --chrome |
| --no-chrome | Disable Chrome integration | claude --no-chrome |
| --channels NEW | Allow MCP servers to push messages into your active session (research preview — server must declare claude/channel capability) | claude --channels |
🎛️ Debugging & Logging
| Flag | Description | Example |
|---|---|---|
| --debug | Enable debug mode with optional category filter | claude --debug "api,mcp" |
| --verbose | Verbose logging — full turn-by-turn output | claude --verbose |
| --mcp-debug | Detailed MCP server error information | claude --mcp-debug |
⚙️ Session & Configuration
| Flag | Description | Example |
|---|---|---|
| -n, --name | Set a display name for the session | claude -n "Feature Implementation" |
| --max-budget-usd | Spending cap for API calls (-p mode only) | claude -p --max-budget-usd 10 "query" |
| --max-turns | Limit agentic turns (-p mode only) | claude -p --max-turns 3 "query" |
| --settings | Load settings from JSON file or inline JSON string | claude --settings ./settings.json |
| --setting-sources | Comma-separated sources: user, project, local | claude --setting-sources user,project |
| --init-only | Run initialization hooks then exit without a session | claude --init-only |
| --from-pr NEW | Resume or create a session linked to a GitHub PR URL | claude --from-pr https://github.com/org/repo/pull/42 |
| --maintenance NEW | Trigger maintenance/setup hooks and then exit (useful in CI for pre-flight setup) | claude --maintenance |
🔧 Miscellaneous CLI Flags
| Flag | Description | Example |
|---|---|---|
| --betas | Beta headers for API requests (API key users only) | claude --betas interleaved-thinking |
| --disable-slash-commands | Disable all slash commands / skills for this session | claude --disable-slash-commands |
| --sandbox NEW | Launch in an isolated sandbox environment with Tab key tab-switching | claude --sandbox |
| --bare NEW 2.1.81 | Scripted -p mode: skips hooks, LSP, plugin sync, and skill directory walks. Requires ANTHROPIC_API_KEY or apiKeyHelper via --settings (OAuth/keychain auth disabled). Auto-memory fully disabled. | claude -p --bare "query" |
Complete Environment Variables Reference
🌍 All Variables
🔑 Authentication & API
| Variable | Purpose | Example Value |
|---|---|---|
| ANTHROPIC_API_KEY | Primary API key; overrides subscription login | sk-ant-xxx... |
| ANTHROPIC_AUTH_TOKEN | Custom Authorization header value (Bearer prefix auto-added) | your-token |
| ANTHROPIC_BETAS | Comma-separated Anthropic beta headers sent with every request (env alternative to --betas flag; applies to all model tiers including Haiku) | interleaved-thinking-2025-05-14 |
| ANTHROPIC_BASE_URL | Route requests via proxy or custom gateway | https://api.example.com |
| ANTHROPIC_CUSTOM_HEADERS | Newline-separated custom HTTP headers (Name: Value) | X-Custom: value |
| ANTHROPIC_FOUNDRY_API_KEY | API key for Microsoft Azure Foundry | xxx... |
| ANTHROPIC_FOUNDRY_BASE_URL | Full base URL for Azure Foundry resource | https://foundry.azure.com |
| ANTHROPIC_FOUNDRY_RESOURCE | Azure Foundry resource name | my-resource |
| ANTHROPIC_FOUNDRY_DEPLOYMENT NEW | Azure Foundry deployment name (separate from resource name) | my-deployment |
| CLAUDE_CODE_CLIENT_CERT | Path to client certificate for mTLS | /path/to/cert.pem |
| CLAUDE_CODE_CLIENT_KEY | Path to private key for mTLS | /path/to/key.pem |
| CLAUDE_CODE_CLIENT_KEY_PASSPHRASE | Passphrase for encrypted mTLS client key | your-passphrase |
| AWS_BEARER_TOKEN_BEDROCK | Bedrock API key / bearer token for authentication | xxx... |
☁️ Cloud Provider Configuration
| Variable | Purpose | Value/Notes |
|---|---|---|
| CLAUDE_CODE_USE_BEDROCK | Route through AWS Bedrock | 1 or true |
| CLAUDE_CODE_USE_VERTEX | Route through Google Vertex AI | 1 or true |
| CLAUDE_CODE_USE_FOUNDRY | Route through Microsoft Azure Foundry | 1 or true |
| CLAUDE_CODE_SKIP_BEDROCK_AUTH | Skip AWS credential check (e.g., when using an LLM gateway) | 1 |
| CLAUDE_CODE_SKIP_VERTEX_AUTH | Skip Google Cloud credential check | 1 |
| CLAUDE_CODE_SKIP_FOUNDRY_AUTH | Skip Azure credential check | 1 |
| ANTHROPIC_BEDROCK_BASE_URL | Proxy URL for Bedrock requests (overrides default Bedrock endpoint) | https://your-proxy-url |
| ANTHROPIC_VERTEX_BASE_URL | Proxy URL for Vertex AI requests | https://your-proxy-url |
| CLOUD_ML_REGION | Google Cloud region for Vertex AI | us-east5 |
| ANTHROPIC_VERTEX_PROJECT_ID | Google Cloud project ID for Vertex AI | my-project-id |
| ANTHROPIC_SMALL_FAST_MODEL_AWS_REGION | Override AWS region for Haiku model on Bedrock | us-west-2 |
| VERTEX_REGION_CLAUDE_3_5_HAIKU | Override Vertex AI region for Haiku model | us-central1 |
🤖 Model Configuration
| Variable | Purpose | Values/Examples |
|---|---|---|
| ANTHROPIC_MODEL | Override default model for all requests | claude-opus-4-6, claude-sonnet-4-6 |
| ANTHROPIC_DEFAULT_SONNET_MODEL | Full model ID the sonnet alias resolves to | claude-sonnet-4-6 |
| ANTHROPIC_DEFAULT_HAIKU_MODEL | Full model ID the haiku alias resolves to | claude-haiku-4-5-20251001 |
| ANTHROPIC_DEFAULT_OPUS_MODEL | Full model ID the opus alias resolves to | claude-opus-4-6 |
| ANTHROPIC_SMALL_FAST_MODEL | DEPRECATED Haiku-class model for background tasks — use specific vars | Use ANTHROPIC_DEFAULT_HAIKU_MODEL |
| CLAUDE_CODE_SUBAGENT_MODEL | Model used specifically for subagents | claude-sonnet-4-6 |
| CLAUDE_CODE_EFFORT_LEVEL | Default thinking effort (Opus 4.6 and Sonnet 4.6 only). Default for Max/Team is now medium as of March 2026 | low | medium | high | max | auto |
| CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING | Disable adaptive reasoning; revert to fixed MAX_THINKING_TOKENS budget | 1 |
| MAX_THINKING_TOKENS | Override extended thinking token budget; 0 disables thinking | 10000 or 0 |
| CLAUDE_CODE_MAX_OUTPUT_TOKENS | Override max output tokens per request (Opus 4.6 default is 64K, upper bound 128K) | 4096 |
| CLAUDE_CODE_DISABLE_1M_CONTEXT | Disable 1M context window; removes 1M variants from /model picker | 1 |
| ANTHROPIC_CUSTOM_MODEL_OPTION | Adds custom entry at bottom of /model picker (skips ID validation) | my-gateway/claude-opus-4-6 |
| ANTHROPIC_CUSTOM_MODEL_OPTION_NAME | Display name for the custom model picker entry (optional) | Opus via Gateway |
| ANTHROPIC_CUSTOM_MODEL_OPTION_DESCRIPTION | Display description for the custom model picker entry (optional) | Custom optimized model |
🧑🤝🧑 Agent Teams (Experimental)
🧪 Experimental Features
Enables parallel AI agent workflows. Disabled by default.
| Variable | Purpose | Value/Notes |
|---|---|---|
| CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS | Enable agent teams feature | 1 or true |
| CLAUDE_CODE_TEAM_NAME | Name of the agent team — auto-set by Claude Code for teammates | my-team |
| CLAUDE_CODE_PLAN_MODE_REQUIRED | Auto-set on teammates requiring plan approval (read-only) | true (auto-managed) |
📌 Enable Teams:
CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 claude --teammate-mode in-process
🔧 Bash & Shell Behavior
| Variable | Purpose | Default / Examples |
|---|---|---|
| BASH_DEFAULT_TIMEOUT_MS | Default timeout for long-running bash commands | 30000 ms |
| BASH_MAX_TIMEOUT_MS | Max timeout the model can request for a bash command | 300000 ms |
| BASH_MAX_OUTPUT_LENGTH | Max characters in bash output before middle-truncation | 500000 |
| CLAUDE_BASH_MAINTAIN_PROJECT_WORKING_DIR | Return to original working dir after each Bash command | 1 |
| CLAUDE_CODE_SHELL | Override automatic shell detection | bash | zsh | sh |
| CLAUDE_CODE_SHELL_PREFIX | Command prefix wrapping all bash commands (logging / auditing) | time or strace |
| CLAUDE_ENV_FILE | Shell script sourced before each Bash command (activate virtualenv, load .env, etc.) | venv/bin/activate |
| CLAUDECODE | Auto-set to 1 inside Claude Code-spawned shells — use to detect context in scripts | 1 (auto-set) |
🧠 Memory, Context & Compaction
| Variable | Purpose | Values / Defaults |
|---|---|---|
| CLAUDE_CODE_DISABLE_AUTO_MEMORY | Disable automatic memory creation and loading | 1 |
| CLAUDE_AUTOCOMPACT_PCT_OVERRIDE | Context fill % at which auto-compaction triggers | 95 (default) |
| CLAUDE_CODE_AUTO_COMPACT_WINDOW | Token capacity used for auto-compaction calculations | Auto-calculated |
| CLAUDE_CODE_ADDITIONAL_DIRECTORIES_CLAUDE_MD | Also load CLAUDE.md from --add-dir directories | 1 |
| CLAUDE_CODE_FILE_READ_MAX_OUTPUT_TOKENS | Override default token limit for file read operations | Model default |
🔌 MCP (Model Context Protocol)
| Variable | Purpose | Values / Examples |
|---|---|---|
| MCP_TIMEOUT | Timeout (ms) for MCP server startup | 30000 ms |
| MCP_TOOL_TIMEOUT | Timeout (ms) for individual MCP tool call execution | 60000 ms |
| MAX_MCP_OUTPUT_TOKENS | Max tokens in MCP tool responses (warning shown above 10K by default) | 50000 |
| ENABLE_CLAUDEAI_MCP_SERVERS | Control claude.ai-managed MCP servers; set false to disable | false to disable |
| ENABLE_TOOL_SEARCH | MCP tool search mode. Auto-mode is now the default since v2.1.80; defers tools that exceed 10% of context | true | auto | auto:N | false |
| MCP_CLIENT_SECRET | OAuth 2.0 client secret for MCP server auth (stored in keychain, not config file) | your-secret |
| MCP_OAUTH_CALLBACK_PORT | Fixed port for OAuth callback URI (use when server has pre-registered redirect URIs) | 8888 |
📋 Tasks, Sessions & Scheduling
| Variable | Purpose | Values |
|---|---|---|
| CLAUDE_CODE_ENABLE_TASKS | Enable task tracking in non-interactive (-p) mode | true |
| CLAUDE_CODE_TASK_LIST_ID | Share a task list across multiple Claude Code instances | task-id-123 |
| CLAUDE_CODE_DISABLE_CRON | Disable scheduled cron jobs and /loop immediately | 1 |
| CLAUDE_CODE_DISABLE_BACKGROUND_TASKS | Disable background task execution and Ctrl+B shortcut | 1 |
| CLAUDE_CODE_EXIT_AFTER_STOP_DELAY | Auto-exit after N ms of idle in SDK/print mode | 5000 ms |
| CLAUDE_CODE_SESSIONEND_HOOKS_TIMEOUT_MS | Max time (ms) for SessionEnd hooks to complete | 1500 ms |
📦 Plugins & Updates
| Variable | Purpose | Values |
|---|---|---|
| DISABLE_AUTOUPDATER | Disable automatic background updates | 1 |
| FORCE_AUTOUPDATE_PLUGINS | Force plugin auto-updates even when autoupdater is disabled | 1 |
| CLAUDE_CODE_PLUGIN_GIT_TIMEOUT_MS | Timeout (ms) for git ops during plugin install/update | 120000 (default) |
| CLAUDE_CODE_PLUGIN_SEED_DIR | Read-only plugin seed directory paths for container images (: on Unix, ; on Windows) | /path/to/plugins |
🚫 Disabling Features
| Variable | Purpose | Value |
|---|---|---|
| CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC | Disables autoupdater, feedback, error reporting, and telemetry — all at once | 1 |
| DISABLE_TELEMETRY | Opt out of Statsig analytics/telemetry | 1 |
| DISABLE_ERROR_REPORTING | Opt out of Sentry crash/error reporting | 1 |
| DISABLE_FEEDBACK_COMMAND | Disable the /feedback slash command | 1 |
| DISABLE_BUG_COMMAND | Alias for DISABLE_FEEDBACK_COMMAND | 1 |
| DISABLE_COST_WARNINGS | Disable cost warning messages during sessions | 1 |
| DISABLE_INSTALLATION_CHECKS | Suppress installation health warnings at startup | 1 |
| DISABLE_PROMPT_CACHING | Disable prompt caching globally across all model tiers | 1 |
| DISABLE_PROMPT_CACHING_HAIKU | Disable prompt caching for Haiku tier only | 1 |
| DISABLE_PROMPT_CACHING_SONNET | Disable prompt caching for Sonnet tier only | 1 |
| DISABLE_PROMPT_CACHING_OPUS | Disable prompt caching for Opus tier only | 1 |
| CLAUDE_CODE_DISABLE_FAST_MODE | Disable fast mode globally | 1 |
| CLAUDE_CODE_DISABLE_TERMINAL_TITLE | Disable automatic terminal title/tab updates | 1 |
| CLAUDE_CODE_DISABLE_GIT_INSTRUCTIONS | Remove built-in git/commit instructions from system prompt | 1 |
| CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS | Strip anthropic-beta headers (for proxies that reject unknown headers) | 1 |
| CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY | Disable "How is Claude doing?" end-of-session surveys | 1 |
| CLAUDE_CODE_SUBPROCESS_ENV_SCRUB NEW 2.1.83 | Strip all Anthropic and cloud-provider credentials from subprocess/child-process environments (prevents credential leakage into spawned shells) | 1 |
🌐 Network & Proxy
| Variable | Purpose | Example Value |
|---|---|---|
| HTTP_PROXY | HTTP proxy server URL | http://proxy.example.com:8080 |
| HTTPS_PROXY | HTTPS proxy server URL | https://proxy.example.com:8080 |
| NO_PROXY | Comma-separated domains/IPs to bypass proxy | localhost,127.0.0.1,.internal.com |
| CLAUDE_CODE_PROXY_RESOLVES_HOSTS | Let the proxy handle DNS resolution instead of the client | 1 |
| CLAUDE_CODE_SKIP_FAST_MODE_NETWORK_ERRORS | Allow fast mode even when org status check fails due to network error | 1 |
⚙️ Miscellaneous / Advanced
| Variable | Purpose | Example Value |
|---|---|---|
| CLAUDE_CONFIG_DIR | Override default config/data directory (default: ~/.claude) | ~/.my-claude-config |
| CLAUDE_CODE_TMPDIR | Override internal temp directory (Claude Code appends /claude/ to the path) | /tmp/my-claude |
| CLAUDE_CODE_SIMPLE | Minimal mode: Bash + file tools only, no MCP/hooks/CLAUDE.md loading | 1 |
| CLAUDE_CODE_NEW_INIT | Enable interactive /init onboarding flow | true |
| CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION | Show/hide grayed-out prompt suggestions; set false to hide | false |
| CLAUDE_CODE_ENABLE_TELEMETRY | Enable OpenTelemetry data collection for enterprise monitoring | 1 |
| CLAUDE_CODE_OTEL_HEADERS_HELPER_DEBOUNCE_MS | Interval (ms) for refreshing dynamic OTel authentication headers | 1740000 (29 min) |
| CLAUDE_CODE_API_KEY_HELPER_TTL_MS | Credentials refresh interval for apiKeyHelper scripts | Auto-managed |
| SLASH_COMMAND_TOOL_CHAR_BUDGET | Override character budget for slash command / skill metadata loading | 5000 |
| USE_BUILTIN_RIPGREP | Bundled ripgrep is used by default; set 0 to use system ripgrep instead | 0 to use system |
| IS_DEMO | Demo mode: hides email/org, skips onboarding | 1 |
| CLAUDE_CODE_HIDE_ACCOUNT_INFO NEW | Hide email and organization from the UI (useful for streaming/recording sessions) | 1 |
| CLAUDE_CODE_ACCOUNT_UUID | SDK callers provide account UUID synchronously at startup | UUID format |
| CLAUDE_CODE_USER_EMAIL | User email for SDK / headless mode | user@example.com |
| CLAUDE_CODE_ORGANIZATION_UUID | Organization UUID for SDK / headless mode | UUID format |
| CLAUDE_CODE_ENABLE_REMOTE_CONTROL_SERVER NEW | Enable remote control server for IDE/web session bridging | 1 |
⚡ Slash Commands (In-Session)
💬 Interactive Mode Only
Type / anywhere in the prompt — not just at the start — to trigger autocomplete. Custom commands in .claude/commands/ appear namespaced by subdirectory (e.g. /frontend:component).
/help
Show all commands including custom skills and MCP server commands
/clear
Clear conversation history and free context. Aliases: /reset, /new
/exit
Exit Claude Code. Alias: /quit
/rename [name]
Rename the current session title
/status
Show version, model, account, and connectivity status
/release-notes
View the latest release notes at any time
/model [model]
Select or change AI model. Use arrow keys to adjust effort when a supported model is active
/effort [level]
Set effort: low | medium | high | max | auto. Shows what auto currently resolves to
/fast [on|off]
Toggle fast mode (speed-optimized Opus 4.6). Note: prior context is re-billed at fast rates when toggled mid-session
/context
Visualize context usage with optimization suggestions. Skills shown as a separate category
/compact [focus]
Compact conversation, optionally retaining focus (e.g., /compact retain the error handling patterns). Use when context > ~80%
/memory
Edit CLAUDE.md, manage auto-memory, view and edit all imported memory files
/init
Initialize project with an interactive CLAUDE.md guide
/add-dir <path>
Add a new working directory to the current session
/config
Open the tabbed Settings interface. Alias: /settings
/permissions
View/update tool permissions. Press / inside to filter. Alias: /allowed-tools
/keybindings
Create or edit ~/.claude/keybindings.json. Changes take effect immediately
/vim
Enable Vim key bindings for prompt input (also configurable in /config)
/plan [description]
Enter plan mode. Pass a description to start immediately (e.g., /plan fix the auth bug)
/output-style [style] DEPRECATED
Deprecated in v2.1.81 — use /config instead. Output style is now fixed at session start for better prompt caching.
/loop [interval] [prompt]
Run a prompt or command on a recurring interval (e.g., /loop 5m check deploy status)
/branch [name]
Fork the current conversation to explore a different path. Alias: /fork
/export [filename]
Export the full conversation as plain text
/copy [N]
Copy last assistant response to clipboard. Press w to write to file instead (useful over SSH)
/btw <question>
Ask a side question without adding it to the main conversation context
/diff
Open the interactive diff viewer for uncommitted changes
/mcp
Manage MCP server connections and OAuth authentication
/plugin
Install, update, and remove Claude Code plugins
/doctor
Diagnose Claude Code installation; validates permission rule syntax and suggests fixes
/debug
Ask Claude to help troubleshoot the current session directly
/stats
Visualize daily usage, session history, and model preferences
/login
Sign in to your Anthropic account (subscription or API Console)
/logout
Sign out from your current account
/feedback
Submit feedback about Claude Code. Alias: /bug
/tasks
List and manage foreground and background tasks. Alias: /bashes
/hooks
View hook configurations for PreToolUse, PostToolUse, and Stop events
/ide
Manage VS Code / JetBrains integrations and show connection status
/color [color]
Set prompt bar color: red, blue, green, yellow, purple, orange, pink, cyan, default
/theme
Open theme selector to change Claude Code's terminal color theme
/remote-control NEW
Bridge terminal session to claude.ai/code — control from any browser or phone
/teleport NEW
Pull a running web/iOS session into your local terminal
/remote-env NEW
Configure environment for remote sessions (claude.ai subscribers)
/desktop NEW
Hand off terminal session to the Desktop app for visual diff review
/sandbox NEW
Launch isolated sandbox environment. Tab / arrow keys switch between tabs
/voice NEW
Toggle voice mode on or off
/terminal NEW
Open an integrated terminal pane within the session
/simplify NEW
Ask Claude to simplify or shorten its last response (built-in bundled command)
/batch <commands> NEW
Run multiple slash commands in a single turn (bundled command)
/reload-plugins NEW
Hot-reload all plugins without restarting Claude Code
/usage NEW
Show token and cost usage summary for the current session
/extra-usage NEW
Detailed breakdown of fast-mode vs normal-mode usage and costs
/resume [id] NEW
Resume a previous session from within an active session (in-session resume)
/security-review NEW
Run a focused security review of the current codebase or recent changes
/tag <name> NEW
Save a named snapshot/tag of the current conversation for later reference
/claude-api NEW
Explore and call Anthropic API endpoints directly from within a session
🪝 Hook Events
⚡ Automation Triggers
Hooks run shell commands or HTTP endpoints at key lifecycle points. Configure in ~/.claude/settings.json under the hooks key. Hooks can be "type": "command" (default) or "type": "http" (POST to a URL). PreToolUse hooks can block tool use by exiting non-zero; PostToolUse hooks receive the result.
Standard Hook Events
| Event | When it fires | Notes |
|---|---|---|
| PreToolUse | Before any tool call executes | Exit non-zero to block the tool. Receives tool name + input JSON on stdin. Filter by matcher field to target specific tools. |
| PostToolUse | After a tool call completes | Receives tool name, input, and output on stdin. Use for logging, linting, or side effects. |
| Stop | When the model stops generating (turn ends) | Runs after each complete assistant turn. |
| StopFailure | When a turn ends in an error/failure state | Use for alerting or rollback automation. |
| PostCompact | After context compaction completes | Useful for reloading context-sensitive state after a /compact. |
| InstructionsLoaded | After all CLAUDE.md files and memory are loaded | Good for injecting dynamic system prompt additions. |
| SessionStart | At the start of a new session | Runs once when Claude Code starts. |
| SessionEnd | When the session ends (exit or /clear) | Timeout controlled by CLAUDE_CODE_SESSIONEND_HOOKS_TIMEOUT_MS. |
| CwdChanged NEW 2.1.83 | Whenever the working directory changes | Ideal for direnv-style automatic environment loading on cd. |
| FileChanged NEW 2.1.83 | When a file is written by a tool | Receives the changed file path. Use for auto-formatting, linting, or reloading. |
| ConfigChange | When settings/config files are reloaded | Fires after dynamic config updates. |
| TeammateIdle | When a teammate agent becomes idle | Available when agent teams are enabled. |
| TaskCompleted | When a background task finishes | Fires on both success and failure of background tasks. |
| Elicitation | When an MCP server sends an elicitation request (needs user input) | Allows custom handling of MCP server prompts. |
| ElicitationResult | After user responds to an MCP elicitation | Receives the user's response for logging or processing. |
Hook Configuration Examples
# ~/.claude/settings.json — command hook (auto-format on file write)
{
"hooks": {
"FileChanged": [
{
"matcher": ".*\\.py$",
"hooks": [{ "type": "command", "command": "black $CLAUDE_FILE_PATHS" }]
}
],
"PreToolUse": [
{
"matcher": "Bash",
"hooks": [{ "type": "command", "command": "echo \"Running: $TOOL_INPUT_COMMAND\" >> ~/audit.log" }]
}
],
"PostToolUse": [
{
"matcher": "Write",
"hooks": [{ "type": "command", "command": "eslint --fix $CLAUDE_FILE_PATHS" }]
}
],
"CwdChanged": [
{
"hooks": [{ "type": "command", "command": "direnv allow && direnv export bash" }]
}
]
}
}
# HTTP hook — send POST to a webhook on every stop
{
"hooks": {
"Stop": [
{
"hooks": [{
"type": "http",
"url": "https://my-server.example.com/claude-webhook",
"headers": { "Authorization": "Bearer my-token" }
}]
}
]
}
}
Hook Environment Variables (available inside hook commands)
| Variable | Available in | Description |
|---|---|---|
| TOOL_NAME | PreToolUse, PostToolUse | Name of the tool being called (e.g., Bash, Write) |
| TOOL_INPUT_COMMAND | PreToolUse (Bash) | The exact command string being run |
| CLAUDE_FILE_PATHS | FileChanged, PostToolUse (file tools) | Space-separated list of affected file paths |
| CLAUDE_SESSION_ID | All hooks | Unique ID for the current session |
| CLAUDECODE | All hooks & subshells | Always 1 — use to detect you're inside Claude Code |
🔣 Variable Substitutions
📝 Custom Commands & Prompts
These placeholders are expanded when Claude Code executes custom slash commands (files in .claude/commands/) and in system prompt templates. Arguments passed to a command are available as $ARGUMENTS or individual positional vars.
| Variable | Expands to | Example Use |
|---|---|---|
| $ARGUMENTS | Full argument string passed to the custom command | /mycommand fix the auth flow → $ARGUMENTS = fix the auth flow |
| $1, $2, ... | Positional arguments (word-split) | /mycommand foo bar → $1=foo, $2=bar |
| ${CLAUDE_SKILL_DIR} | Absolute path to the directory containing the current skill/command file | Useful for loading sibling files relative to the command |
| ${CLAUDE_SESSION_ID} | Unique session identifier (UUID) | Use in logs or filenames to correlate with hook output |
| ${CLAUDE_CWD} | Current working directory when the command was invoked | Useful for path-relative operations in commands |
| !`shell command` | Output of a shell command, injected into the prompt | !`cat README.md` — injects file contents at invocation time |
Custom Command Frontmatter
Custom commands (Markdown files in .claude/commands/) support YAML frontmatter to control behavior:
---
description: "Generate a component with tests"
effort: high # low | medium | high | max | auto
model: opus # override model for this command
disableMemory: true # skip loading CLAUDE.md memory for this command
allowedTools: # restrict tools available during this command
- Read
- Write
- Bash
---
Create a $ARGUMENTS component with unit tests. Use $1 as the component name.
Custom Command Namespacing
Files in subdirectories of .claude/commands/ are namespaced by folder:
.claude/commands/
├── deploy.md → /deploy
├── frontend/
│ ├── component.md → /frontend:component
│ └── test.md → /frontend:test
└── backend/
└── migrate.md → /backend:migrate
⚙️ Settings.json Reference
🗂️ Configuration Files
Settings are layered in this priority order (highest wins): local (.claude/settings.local.json) → project (.claude/settings.json) → user (~/.claude/settings.json) → managed (/etc/claude-code/managed-settings.json or managed-settings.d/). Use --setting-sources to control which layers are loaded.
Model & Behavior Settings
| Key | Purpose | Values |
|---|---|---|
| model | Default model for the session | "claude-opus-4-6", "sonnet", "haiku" |
| modelOverrides | Per-tier model overrides (e.g., override just the haiku alias) | {"haiku": "claude-haiku-4-5-20251001"} |
| effortLevel | Default effort level | "low" | "medium" | "high" | "max" | "auto" |
| includeGitInstructions | Include built-in git/commit instructions in system prompt (default: true) | false to disable |
| outputStyle | Response output style — set at session start for stable prompt caching | "auto" | "concise" | "verbose" |
| showTurnDuration | Show elapsed time for each assistant turn in the UI | true | false |
| feedbackSurveyRate | Frequency of end-of-session feedback surveys (0 = never, 1 = always) | 0–1 (float) |
| reducedMotion | Reduce or disable animations in the UI | true |
Memory & Context Settings
| Key | Purpose | Values |
|---|---|---|
| autoMemoryDirectory | Override where auto-generated memory files are stored | "~/.claude/memory" (path string) |
| clearContextOnStart | Start each new session with a fresh context (no continuation) | true | false |
| plansDirectory | Custom directory for plan mode output files | ".claude/plans" |
Sandbox & Security Settings
| Key | Purpose | Values |
|---|---|---|
| sandbox | Enable or configure sandbox for all sessions | true | false | object |
| sandbox.failIfUnavailable NEW 2.1.83 | Exit with error if sandbox is enabled but cannot start (instead of silently falling back) | true |
| sandbox.enableWeakerNetworkIsolation | Allow limited outbound network access inside sandbox (weaker isolation) | true |
| sandbox.allowRead | Additional paths sandbox is allowed to read from (array of glob strings) | ["/home/user/data/**"] |
| disableDeepLinkRegistration NEW 2.1.83 | Prevent Claude Code from registering the claude-cli:// protocol handler on the system | true |
LSP & IDE Settings
| Key | Purpose | Values |
|---|---|---|
| lspStartupTimeout | Milliseconds to wait for LSP server startup before giving up | 10000 (default) |
| preferredNotificationsFormat | Notification format for IDE plugin events | "json" | "text" |
Plugin & Skill Settings
| Key | Purpose | Values |
|---|---|---|
| pluginMarketplaceSource | URL of the plugin marketplace index (for custom/enterprise marketplaces) | "https://my-registry.example.com/index.json" |
| spinner | Customize the loading spinner style in the terminal UI | "dots" | "line" | "minimal" |
| chat:newline | Keybinding action to insert a newline in multi-line input mode | "shift+enter" (configurable via /keybindings) |
Voice Settings
| Key | Purpose | Values |
|---|---|---|
| voice.enabled | Enable voice mode in the session | true | false |
| voice.language | Language/locale for voice recognition | "en-US", "ja-JP", etc. |
| voice.autoSubmit | Auto-submit prompt after silence is detected | true | false |
Worktree Settings
| Key | Purpose | Values |
|---|---|---|
| worktree.sparsePaths | Limit git worktree sparse checkout to specific paths (array) | ["src/", "tests/"] |
Settings.json Full Example
// ~/.claude/settings.json
{
"model": "claude-opus-4-6",
"effortLevel": "medium",
"includeGitInstructions": true,
"autoMemoryDirectory": "~/.claude/memory",
"sandbox": {
"failIfUnavailable": true,
"enableWeakerNetworkIsolation": false,
"allowRead": ["/home/user/datasets/**"]
},
"disableDeepLinkRegistration": false,
"voice": { "enabled": false },
"lspStartupTimeout": 10000,
"showTurnDuration": true,
"feedbackSurveyRate": 0,
"reducedMotion": false,
"env": {
"ANTHROPIC_API_KEY": "sk-ant-xxx...",
"CLAUDE_CODE_SHELL": "zsh"
},
"permissions": {
"allow": ["Bash(git *)", "Read(**)", "Write(src/**)"],
"deny": ["Bash(rm -rf *)"]
},
"hooks": {
"FileChanged": [
{ "hooks": [{ "type": "command", "command": "prettier --write $CLAUDE_FILE_PATHS" }] }
]
}
}
🏢 Managed Settings (Enterprise)
🔒 Admin / Fleet Deployment
Managed settings are applied system-wide and cannot be overridden by users or project settings. They are read from /etc/claude-code/managed-settings.json (Linux/macOS) or the equivalent Windows path. New in v2.1.83: place fragment files in /etc/claude-code/managed-settings.d/ — files are merged alphabetically, allowing independent teams to deploy separate policy fragments.
Key Managed Settings
| Key | Purpose | Values |
|---|---|---|
| disableAutoMode NEW 2.1.82 | Prevent users from enabling auto mode (research preview) organisation-wide | true |
| apiKeyHelper | Shell command that outputs a fresh API key (used for token rotation) | "python3 /opt/get-key.py" |
| disableTelemetry | Disable all telemetry and analytics organisation-wide | true |
| disableErrorReporting | Disable Sentry crash reporting organisation-wide | true |
| disableAutoUpdater | Prevent automatic background updates | true |
| disableFeedbackCommand | Remove /feedback command for all users | true |
| disableDeepLinkRegistration | Prevent claude-cli:// protocol handler registration | true |
| permissions.allow | Org-wide allowed tool patterns (merged with user/project allows) | ["Bash(git *)", "Read(**)"] |
| permissions.deny | Org-wide denied tool patterns — cannot be overridden by users | ["Bash(curl *)", "Bash(wget *)"] |
| env | Environment variables applied to all sessions fleet-wide | {"HTTPS_PROXY": "http://corp-proxy:8080"} |
managed-settings.d/ Drop-in Fragments NEW 2.1.83
# Deploy individual policy fragments — merged alphabetically
/etc/claude-code/managed-settings.d/
├── 00-base.json # base org policy
├── 10-security-team.json # security team overrides
└── 20-ml-platform.json # ML platform team additions
# Example fragment: /etc/claude-code/managed-settings.d/10-security-team.json
{
"permissions": {
"deny": ["Bash(curl *)", "Bash(wget *)"]
},
"disableTelemetry": true
}
🎯 Quick Reference Matrix
📊 Common Patterns
| Task | Command / Flag / Variable |
|---|---|
| Start interactive session | claude |
| Continue previous conversation | claude -c or claude --continue |
| Quick non-interactive query | claude -p "query" |
| Change model to Opus | claude --model claude-opus-4-6 |
| Set effort level | claude --effort high or CLAUDE_CODE_EFFORT_LEVEL=high |
| Enable experimental agent teams | CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 |
| Use AWS Bedrock | CLAUDE_CODE_USE_BEDROCK=1 + AWS_REGION=us-east-1 |
| Use Google Vertex AI | CLAUDE_CODE_USE_VERTEX=1 + CLOUD_ML_REGION=us-east5 + ANTHROPIC_VERTEX_PROJECT_ID=my-proj |
| Skip permission prompts (devcontainer only) | claude --dangerously-skip-permissions |
| Enter plan mode at launch | claude --permission-mode plan |
| Add multiple directories | claude --add-dir ../apps ../lib |
| Append to system prompt | claude --append-system-prompt "Always use TypeScript" |
| JSON output in -p mode | claude -p --output-format json "query" |
| Restrict tools to safe set | claude --tools "Bash,Read,Grep,Edit" |
| Disable all non-essential traffic | CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1 |
| Load MCP servers | claude --mcp-config ./mcp.json |
| Enable MCP push channels | claude --channels |
| Corporate HTTPS proxy | HTTPS_PROXY=http://proxy:8080 claude |
| Custom shell | CLAUDE_CODE_SHELL=zsh claude |
| Auto-activate virtualenv | CLAUDE_ENV_FILE=.venv/bin/activate claude |
| Fork session for parallel work | claude --fork-session |
| Isolated git worktree | claude --worktree |
| Disable extended thinking | MAX_THINKING_TOKENS=0 |
| Disable 1M context window | CLAUDE_CODE_DISABLE_1M_CONTEXT=1 |
| Bridge session to browser/phone | claude --remote-control or in-session /remote-control |
| Pull web session to terminal | In-session: /teleport |
| Launch sandbox environment | claude --sandbox |
| Run recurring check every 5 min | In-session: /loop 5m check deploy status |
| Hide account info from UI | CLAUDE_CODE_HIDE_ACCOUNT_INFO=1 |
| Use system ripgrep | USE_BUILTIN_RIPGREP=0 |
| Debug MCP issues | claude --mcp-debug --verbose |
| Fast scripted automation (no hooks/LSP/plugins) | claude -p --bare "query" |
| Enable auto mode (research preview) | claude --enable-auto-mode |
| Strip credentials from subprocesses | CLAUDE_CODE_SUBPROCESS_ENV_SCRUB=1 |
| Resume from GitHub PR | claude --from-pr https://github.com/org/repo/pull/42 |
| List configured agents | claude agents |
| Manage MCP servers | claude mcp list / claude mcp add |
| Transcript search (in-session) | Press / then n/N to navigate |
📥 Installation & Setup
🚀 Getting Started
⚠️ npm install is deprecated.
npm install -g @anthropic-ai/claude-code is no longer the recommended method. Use native installers below. Native installs auto-update in the background.
macOS & Linux — Install Script (Recommended)
# Latest version
curl -fsSL https://claude.ai/install.sh | bash
# Specific version
curl -fsSL https://claude.ai/install.sh | bash -s 2.1.83
macOS — Homebrew
brew install --cask claude-code
# Note: Homebrew does NOT auto-update. Run periodically:
brew upgrade claude-code
Windows — WinGet (Recommended for Windows)
# Requires Git for Windows installed first
winget install Anthropic.ClaudeCode
# Note: WinGet does NOT auto-update. Run periodically:
winget upgrade Anthropic.ClaudeCode
Windows — PowerShell Script
# Latest
irm https://claude.ai/install.ps1 | iex
# Specific version
& ([scriptblock]::Create((irm https://claude.ai/install.ps1))) 2.1.83
Authentication
# Login with claude.ai subscription
claude auth login
# Login via Anthropic Console / API billing account
claude auth login --console
# Or set API key directly
export ANTHROPIC_API_KEY=sk-ant-xxx...
# Check auth status
claude auth status
# Logout
claude auth logout
Update
# Manual update (native installs also update automatically)
claude update
# Force reinstall (bypasses stale lock files)
claude install --force
npm — Legacy / CI Fallback
# ⚠️ DEPRECATED — use native installers above
npm install -g @anthropic-ai/claude-code@2.1.83
💡 Tips & Best Practices
✨ Pro Tips
🔍 Search: Ctrl+F (Cmd+F on Mac) to find any variable, flag, or command on this page.
⚙️ Environment Variables: Set temporarily by prefixing your command, or permanently in your shell profile. Can also be set under
env in ~/.claude/settings.json for team-wide rollout.🧪 Experimental Features: Agent teams and some v2.1.80 features are experimental — avoid in production without testing.
📝 Context Management: Use /compact when context > ~80% (instant since v2.0.64). Use /clear when switching to an unrelated task entirely.
🔐 Security:
--dangerously-skip-permissions bypasses all safety checks. Use only in a devcontainer with a firewall configured — never on your main machine.💰 Cost: Enabling fast mode mid-session re-bills the entire prior context at fast rates. Use
--max-budget-usd to cap spending in -p mode.🎯 Model Choice: Opus 4.6 (1M context, 64K output) for complex reasoning; Sonnet 4.6 for daily work; Haiku for subagents and cheap searches. Default effort for Max/Team Opus subscribers changed to
medium in March 2026.🆕 v2.1.81 Highlights:
--bare flag for fast deterministic scripted -p automation (skips hooks/LSP/plugins/skills), --channels permission relay (channel servers can forward tool approval prompts to phone), default Opus model on Bedrock/Vertex/Foundry changed to Opus 4.6 (was 4.1), /output-style deprecated (use /config), ANTHROPIC_BETAS env var now correctly applied to all model tiers including Haiku, VS Code spark icon + native MCP management dialog, Remote Control poll rate cut ~300×.🆕 v2.1.82 Highlights:
--enable-auto-mode (research preview) — Claude auto-approves low-risk actions via Sonnet 4.6 classifier, safer middle ground between default and --dangerously-skip-permissions. Configurable multi-line input keybinding (chat:newline). Managed setting disableAutoMode to block auto mode org-wide.🆕 v2.1.83 Highlights:
managed-settings.d/ drop-in directory for enterprise policy fragments, CLAUDE_CODE_SUBPROCESS_ENV_SCRUB to strip credentials from subprocesses, CwdChanged / FileChanged hook events for reactive environment management, sandbox.failIfUnavailable setting, transcript search with / key (n/N to navigate matches).🎓 Real-World Examples
Example 1: Agent Teams (Experimental)
export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1
claude --teammate-mode in-process --permission-mode plan
Example 2: High-Effort Analysis on Opus
claude --model claude-opus-4-6 --effort high \
--append-system-prompt "Focus on security and performance implications"
Example 3: Privacy / Air-Gap Setup
export CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1
export ANTHROPIC_API_KEY=sk-ant-xxx...
claude
Example 4: Enterprise Bedrock with Proxy & Pinned Models
export HTTPS_PROXY=http://corporate-proxy:8080
export CLAUDE_CODE_USE_BEDROCK=1
export AWS_REGION=us-east-1
# Pin exact model versions to prevent breakage on Anthropic updates
export ANTHROPIC_DEFAULT_OPUS_MODEL='us.anthropic.claude-opus-4-6-v1'
export ANTHROPIC_DEFAULT_SONNET_MODEL='us.anthropic.claude-sonnet-4-6'
export ANTHROPIC_DEFAULT_HAIKU_MODEL='us.anthropic.claude-haiku-4-5-20251001-v1:0'
export CLAUDE_CODE_SHELL_PREFIX="time" # Audit command timings
claude
Example 5: CI/CD Integration (Fully Automated)
# Best run inside a devcontainer with firewall
claude -p \
--output-format json \
--max-turns 10 \
--max-budget-usd 5 \
--dangerously-skip-permissions \
--tools "Bash,Read,Grep,Edit,Write" \
"Analyze this PR for security vulnerabilities and output findings" \
> analysis.json
Example 6: Strict MCP-Only Environment
claude \
--mcp-config ./mcp-servers.json \
--strict-mcp-config \
--plugin-dir ./plugins \
--append-system-prompt "Use MCP tools for all external data"
Example 7: Python Virtual Environment Auto-Activation
export CLAUDE_ENV_FILE=.venv/bin/activate
export CLAUDE_CODE_SHELL=bash
claude --add-dir ./backend ./frontend
Example 8: Debug MCP & API Issues
claude --verbose --mcp-debug --debug "api,mcp,hooks" \
--append-system-prompt "Include diagnostic details in responses"
Example 9: Remote Control & MCP Channels (v2.1.80)
# Bridge terminal to browser or phone
claude --remote-control
# Or enable MCP push channels (server must declare claude/channel capability)
claude --channels --mcp-config ./mcp-servers.json
# Or start a session then bridge mid-session
claude
# then type: /remote-control
Example 10: Recurring Monitor with /loop
claude
# then type:
/loop 10m check test suite status and summarize any failures
Example 11: Cross-Surface Workflow
# Start long-running task on claude.ai web
# Later, pull it into your terminal:
claude
# then type: /teleport
# Or hand a terminal session off to the Desktop app for visual diff:
# type: /desktop
Example 12: Bare Scripted Automation (v2.1.81)
# --bare skips hooks, LSP, plugin sync, skill walks — fastest deterministic -p calls
# Requires ANTHROPIC_API_KEY (OAuth/keychain disabled in bare mode)
export ANTHROPIC_API_KEY=sk-ant-xxx...
claude -p --bare \
--output-format json \
--max-turns 5 \
"Summarize all TODO comments in this repo" \
> todos.json
Example 13: Auto Mode (v2.1.82)
# Auto mode — Claude auto-approves low-risk actions (research preview)
# Safer than --dangerously-skip-permissions, more autonomous than default
claude --enable-auto-mode "implement the feature described in TASK.md"
# Disable auto mode org-wide via managed settings:
# /etc/claude-code/managed-settings.json → { "disableAutoMode": true }
Example 14: Managed Settings Drop-in (v2.1.83)
# Deploy policy fragments independently per team
# /etc/claude-code/managed-settings.d/10-security.json
{
"permissions": { "deny": ["Bash(curl *)", "Bash(wget *)"] },
"env": { "CLAUDE_CODE_SUBPROCESS_ENV_SCRUB": "1" }
}