Permissions

The Dashcat connector, and each bridge paired with a node, has a permission for each kind of action. Permissions live on the node, so they hold whatever the AI on the other end is told to do. To choose which nodes an AI app can reach at all, use Connections instead.

Action	Covers	MCP tools
`read`	Reading files and logs, listing the programs Dashcat started and reading their output	`read_file`, `tail_file`, `list_processes`, `process_output`
`screen`	Seeing what is on screen	`screenshot`, `ui_tree`
`run`	Running commands, starting and stopping programs	`run`, `start_process`, `stop_process`
`write`	Creating, replacing or adding to files	`write_file`, `dashcat-bridge push`
`control`	Clicking, typing and operating apps	`ui_action`, `click`, `scroll`, `type_text`, `press_keys`

node_info is always allowed.

Each action is set to one of:

allow: the bridge does it without asking. This is the default for every action, so a new pairing works as it always has.
ask: the node asks the person sitting at it first.
deny: the node refuses, and tells the AI not to look for another way.

Changing permissions

On the node:

dashcat-agent bridges list                       # Each bridge and its permissions
dashcat-agent bridges ask laptop run write       # Ask before commands and file changes
dashcat-agent bridges ask laptop all             # Ask before everything
dashcat-agent bridges deny laptop control        # Never click or type in apps
dashcat-agent bridges allow laptop all           # Back to the default

Name the bridge as bridges list shows it, or by its fingerprint. Changes apply to the next request, also in sessions that are already open.

The Dashcat connector

A node that joined your Dashcat account has permissions for the Dashcat connector too: what your AI may do on it from claude.ai, the Claude apps or any other AI app that uses the connector. They work like a bridge’s and also start with everything allowed. The connector is called connector:

dashcat-agent bridges deny connector control     # The connector may not click or type in apps
dashcat-agent bridges ask connector run          # Ask before the connector runs commands

To keep the connector out of a node entirely, join it with --hosted deny instead. To shut out the connector and every bridge for a while, choose Pause remote access in the tray menu.

Which AI apps can reach which nodes

Each AI app you sign in to the connector shows up under Connections in Dashcat Dashboard, with when it was last used. A new one can work on every node that allows the connector. Use Choose nodes to limit it to some of them, or Remove to shut it out at once; to use it again, sign in from the app once more. Nodes it can’t reach are hidden from it.

The first time a new AI app uses a node, the node’s tray shows a notice, and the activity log names who asked and the app on every request, for example Dashcat connector (@tom via Claude (claude.ai)).

Answering a request

When an action is set to ask, the node shows a message saying which bridge wants to do what, for example “laptop wants to run commands: Run: brew upgrade”. The request then waits in the Dashcat icon’s menu, in the notification area on Windows or the menu bar on macOS, with four choices:

Allow once: run this request.
Allow for an hour: run this and the same kind of request from this bridge for the next hour.
Always allow: set this action to allow for this bridge.
Deny: refuse it.

If nobody answers within 2 minutes, the request is refused. Answers and permission changes are recorded in the activity log. Typed text and field values are shown as their length only, because they can be passwords.

Asking needs the tray icon. A node running without one (over SSH, with --no-tray, or on a machine with no desktop) refuses actions set to ask, and the AI is told how to allow them.