SESSIONS & CONTROL
Use Chrome extensions in a remote browser
Tab view cannot show a toolbar popup. Full browser can. That view also reaches Chrome settings and file dialogs on the host.
Tab and Full browser
| View | What you see |
|---|---|
| Tab | The page. Chrome's toolbar and popups are missing. |
| Full browser | Chrome itself: toolbar, extension popups, side panels, and dialogs. |
Full browser has no address bar of its own. Chrome's is in the picture, so click into it and type there. Your keyboard reaches Chrome only while the pointer is over its window, which is why the view fits Chrome to the whole display when it opens.
Switch it on the browser page after you take control. It needs Linux, a dedicated Xvfb display, ffmpeg, and xdotool. The published 0.11.6 image includes those tools.
Install from the Chrome Web Store
Check extensions are on
New browsers start with extension support on. If it was turned off, stop the browser, choose Enable extensions, and start it again.
Open Full browser
Take control, then choose Full browser. Chrome is fitted to the display for you; Refit Chrome window puts it back if a dialog moves it.
Install the extension
Open the Chrome Web Store in Chrome's address bar and install it there. Permission dialogs show in this view.
Use or leave it
Open the toolbar popup, or Manage extensions for chrome://extensions/. Switch back to Tab for ordinary work.
Let an agent use the popup
Agents cannot turn extension support on. Allow agent control is a second toggle, for the owning agent only. New agent-owned browsers start with it on. Borrowed browsers do not get it. Copies do not inherit it.
When it is off, the agent is told to ask you to open this browser in the dashboard, go to Extensions, turn on Allow agent control, and wait. Do not expect it to retry its way through that gate.
Set TALLYLAMP_AGENT_DESKTOP_DEFAULT=0 to start new agent-owned browsers with it off. Existing browsers keep whatever you already chose. Turning the toggle off still wins. This does not enable extensions.
| Tool | Job |
|---|---|
| tallylamp_desktop_screenshot | See the whole Chrome window, including popups. |
| tallylamp_desktop_action | Click, type, scroll, send a shortcut, or open chrome://extensions/. |
Coordinates are screen pixels, not the downscaled image. Human control stops these tools. Taking control or turning the permission off interrupts in-flight native work.
Let an agent debug an extension
With extensions on, the browser's Chrome DevTools tools reach extensions too. The agent can read an extension's service worker and content-script logs, run code in its service worker, open its pages, and press its toolbar button. These are ordinary MCP tools, so they do not need Allow agent control.
Only the browser's owner can install, reload or uninstall. A borrowed browser refuses those at either access level, because an installed extension keeps running after the loan ends. Linked browsers get none of these tools.
| Tool | Job |
|---|---|
| list_pages | Lists extension service workers alongside pages. |
| evaluate_script | With serviceWorkerId, runs code inside an extension's service worker, with that extension's permissions. |
| list_console_messages | Includes messages from content scripts. |
| navigate_page | Opens an extension's own pages, such as chrome-extension://<id>/popup.html. |
| list_extensions | Lists unpacked extensions. |
| trigger_extension_action | Presses an extension's toolbar button on the selected page. |
| install_extension | Loads an unpacked folder from the server's disk. Refused for a browser on a worker. |
| reload_extension | Loads it again after its files change. |
| uninstall_extension | Removes it. |
Docs behind this guide
Based on Tallylamp's documented behaviour. See the source for API details and release-specific settings.