Tools & MCP commands

/mcp — manage MCP servers

Type: local-jsx
Syntax: /mcp [enable|disable [server-name]]Opens the MCP (Model Context Protocol) management panel. Use this to list, enable, disable, add, or remove MCP servers.Adding a server via the CLI:

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

Flag	Description
`-s, --scope <scope>`	Configuration scope: `local`, `user`, or `project` (default: `local`)
`-t, --transport <type>`	Transport type: `stdio`, `sse`, or `http` (default: `stdio`)
`-e, --env <KEY=value>`	Pass environment variables to the server
`-H, --header <header>`	Set HTTP headers
`--client-id <id>`	OAuth client ID
`--client-secret`	Prompt for OAuth client secret
`--callback-port <port>`	Fixed OAuth callback port
`--xaa`	Enable XAA (SEP-990) authentication

Examples:

# Add a local stdio server
claude mcp add my-tools npx my-mcp-package

# Add an SSE server scoped to the project
claude mcp add my-api https://api.example.com/mcp -s project -t sse

# Disable a server in the current session
/mcp disable my-tools

XAA IdP subcommands (enterprise identity provider integration):

Subcommand	Description
`claude mcp xaa setup`	Configure IdP: `--issuer <url>` `--client-id <id>` `[--client-secret]`
`claude mcp xaa login`	Authenticate with IdP via OIDC browser flow. `[--force]`
`claude mcp xaa show`	Show current IdP config and login status
`claude mcp xaa clear`	Clear IdP config and cached `id_token`

/permissions — manage tool permissions

Aliases: allowed-tools
Type: local-jsx
Syntax: /permissionsOpens the interactive permissions manager, where you can set allow and deny rules for individual tools or tool patterns.Permission rules are stored in settings files (~/.claude/settings.json for global rules, .claude/settings.json for project rules) and support wildcard patterns.

/permissions
# or equivalently:
/allowed-tools

Permission levels:

Level	Behavior
Automatic	Allowed without prompting
Ask once	Prompt once per session and remember
Ask always	Prompt on every invocation
Deny	Block completely

/hooks — configure tool hooks

Type: local-jsx
Syntax: /hooksOpens the hooks configuration panel. Hooks are scripts that run automatically at tool lifecycle events.Supported hook events:

Event	Fires when
`PreToolUse`	Before a tool is called
`PostToolUse`	After a tool returns
`Stop`	When the model stops generating
`SessionStart`	When a session begins
`SessionEnd`	When a session ends (also triggered by `/clear`)

/hooks

/agents — manage agents

Type: local-jsx
Syntax: /agentsOpens the agents management menu. From here you can create, edit, and delete custom agent configurations. Agents are isolated Claude instances used for background tasks, parallel work, and sub-agent delegation.

/agents

/files — list context files

Type: local
Syntax: /filesLists all files currently loaded in the context (the read-file state cache).

/files
# Output: relative paths of all files in context, or "No files in context."

This command is currently restricted to Anthropic-internal builds (USER_TYPE=ant).

/add-dir — add a directory to context

Type: local-jsx
Syntax: /add-dir [<path>]Adds a working directory to the session’s permission context. This lets Claurst read and write files in directories outside the initial working directory.Arguments:

Argument	Description
(none)	Open the interactive directory picker
`<path>`	Add the specified path directly

Behavior:

Expands ~ and resolves symlinks before validating.
Prompts to save the addition to local settings for future sessions.
Updates the bash sandbox configuration.
Handles ENOENT, ENOTDIR, EACCES, and EPERM errors gracefully.

Examples:

/add-dir
/add-dir ~/projects/shared-lib

/skills — list available skills

Type: local-jsx
Syntax: /skillsDisplays all currently loaded skills — custom slash command macros loaded from skill directories, plugins, and bundled skills. Skills appear as additional commands in the command palette.

/skills

/plugin — manage plugins

Aliases: plugins, marketplace
Type: local-jsx
Syntax: /plugin [subcommand] [args]Manages Claude Code plugins: install, uninstall, enable, disable, and browse the marketplace.Subcommands:

Subcommand	Syntax	Description
(none)	`/plugin`	Open the interactive plugin menu
`install` / `i`	`/plugin install [plugin@marketplace]`	Install a plugin
`manage`	`/plugin manage`	Open the manage-plugins UI
`uninstall`	`/plugin uninstall [plugin]`	Uninstall a plugin
`enable`	`/plugin enable [plugin]`	Enable a disabled plugin
`disable`	`/plugin disable [plugin]`	Disable a plugin without removing it
`validate`	`/plugin validate [path]`	Validate a plugin definition file
`marketplace`	`/plugin marketplace [add\|remove\|update\|list] [target]`	Manage registered marketplaces
`help`	`/plugin help`	Show plugin command help

Plugin format: plugin-name@marketplace-name or plugin-name@https://marketplace-urlExamples:

/plugin
/plugin install pr-comments@claude-code-marketplace
/plugin disable my-custom-plugin
/plugin validate ./my-plugin/plugin.json

/reload-plugins — activate pending plugin changes

Type: local
Syntax: /reload-pluginsApplies pending plugin changes in the current session without restarting. Reports a summary of what was reloaded.Example output:

Reloaded: 3 plugins · 5 skills · 1 agent · 2 hooks · 1 plugin MCP server · 0 plugin LSP servers

If there were load errors: 2 errors during load. Run /doctor for details.

/init — initialize project settings

Type: prompt
Syntax: /initInitializes CLAUDE.md (and optionally skills and hooks) for the current repository. This is the recommended first step when starting to use Claurst in a new project.What it sets up (new multi-phase flow when CLAUDE_CODE_NEW_INIT=1):

Phase	Action
1	Ask what to set up (project `CLAUDE.md`, personal `CLAUDE.local.md`, skills, hooks)
2	Explore codebase (manifests, README, CI config, linter config, git worktrees)
3	Fill gaps via interactive questions
4	Write `CLAUDE.md` at project root
5	Write `CLAUDE.local.md` (auto-added to `.gitignore`)
6	Suggest and create skills in `.claude/skills/`
7	Suggest additional optimizations (GitHub CLI, linting, hooks)
8	Summary and next steps (plugin suggestions)

/init

/init-verifiers — create verifier skills

Type: prompt
Syntax: /init-verifiersCreates verifier skill files in .claude/skills/ for automated verification of code changes. Verifiers can be invoked by the Verify agent.Supported verifier types:

Type	Allowed tools
`verifier-playwright`	`Bash(npm:)`, `Bash(yarn:)`, `mcp__playwright__*`, Read, Glob, Grep
`verifier-cli`	Tmux, `Bash(asciinema:*)`, Read, Glob, Grep
`verifier-api`	`Bash(curl:)`, `Bash(http:)`, `Bash(npm:*)`, Read, Glob, Grep

Phases: Auto-detects project type → sets up verification tools → interactive Q&A → generates skills → confirms.

/init-verifiers is currently restricted to Anthropic-internal builds.

/advisor — configure a secondary advisor model

Type: local
Syntax: /advisor [<model>|off|unset]Configures a secondary model that reviews and advises on the primary model’s outputs in real time.Arguments:

Argument	Description
(none)	Show the current advisor setting
`<model>`	Set the advisor model (e.g. `opus`, `claude-opus-4-6`)
`off` / `unset`	Disable the advisor

Examples:

/advisor claude-opus-4-6
/advisor off

/brief — toggle brief-only output mode

Type: local-jsx
Syntax: /briefToggles “brief-only mode.” When enabled, the model routes all user-facing output through the SendUserMessage (Brief) tool; plain text output is hidden. Intended for structured, machine-readable workflows.Gate: Feature-gated via the KAIROS or KAIROS_BRIEF bundle flags plus a GrowthBook config flag. Hidden when not entitled.

/rate-limit-options — show rate-limit recovery options

Type: local-jsx
Syntax: /rate-limit-optionsDisplays options when the rate limit is reached (e.g., upgrade plan, configure extra usage).

This command is hidden — it is triggered automatically from the rate-limit message in the UI, not invoked by users directly.

/insights — generate a usage analysis report

Type: prompt
Syntax: /insightsGenerates an AI-powered report analyzing Claude Code usage sessions. Reads session files from ~/.claude/projects/, extracts patterns, runs summarization via the Opus model, and produces HTML and diff-line reports.

/insights

/tasks — list background tasks

Aliases: bashes
Type: local-jsx
Syntax: /tasksLists and manages running background tasks — both shell command tasks and agent tasks.

/tasks
/bashes

/sandbox — configure bash sandbox settings

Type: local-jsx
Syntax: /sandbox [exclude "command pattern"]Configures sandbox settings for bash command execution. The command description dynamically shows the current sandbox status (enabled/disabled, auto-allow mode, fallback allowed, managed policy).Example:

/sandbox exclude "grep *"

Gate: Hidden on unsupported platforms.

/statusline — configure the status line UI

Type: prompt
Syntax: /statusline [description]Configures the Claurst status line display. Launches a statusline-setup sub-agent that reads your shell PS1 configuration and writes updated settings to ~/.claude/settings.json.

/statusline
/statusline use my existing zsh PS1 theme

/ide — manage IDE integrations

Type: local-jsx
Syntax: /ide [open]Manages IDE integrations (VS Code, JetBrains) and shows the connection status. Use open to launch the integration setup.

/ide
/ide open

/remote-control — start a remote control bridge

Aliases: rc
Type: local-jsx
Syntax: /remote-control [name]Starts or manages a Remote Control bridge session, allowing your terminal session to be controlled from a mobile device or web interface via a QR code.Arguments:

Argument	Description
(none)	Start a bridge session with auto-generated name
`<name>`	Use a custom session name shown in the remote client

Gate: Requires the BRIDGE_MODE feature flag and isBridgeEnabled() to return true.

/remote-control my-session
/rc

/ultraplan — extended multi-agent planning (internal)

Type: local-jsx
Syntax: /ultraplan [seed plan]Runs an extended multi-agent planning session on Claude Code on the web (CCR). The remote session explores the codebase, creates a comprehensive implementation plan, and enters plan-approval mode when done (up to 30 minutes).Gate: Requires the ULTRAPLAN bundle flag. Internal-only (USER_TYPE=ant).

/ultraplan add OAuth2 support to the API

Commands

Tools

Architecture

Tools & MCP commands