An agentic browser harness: an AI side panel & orb in your own logged-in Chrome, Edge, Arc, Brave or even Firefox run by the agent/subscription you already use (Claude Code, Codex, Antigravity). Free, open source, no API key. Also an MCP server.
Background monitoring
A monitor watches a long job in a tab, such as an upload, a build or a deploy, and notifies you when it finishes. The browser does the watching, so no agent spends tokens asking whether the job is done.
watch this upload and tell me when it's done
How a monitor runs
Browsentic finds the progress signal, pins the tab, and watches it in the background while you work elsewhere. It tracks the percentage, extrapolates an ETA from the sample history, notices when progress has stalled, and notifies you on completion.
The watch runs in the browser, in the extension. It needs no further tool calls, and it keeps running if the agent finishes, the MCP client disconnects, or Browsentic Bridge goes away.
What it can watch for
| Condition | Completes when |
|---|---|
element-appears |
An element shows up |
element-vanishes |
An element goes away, usually a spinner |
text-matches |
Page text matches a regular expression |
title-matches |
The tab title matches one |
progress-reaches |
A progress bar hits a threshold (100 by default) |
page_findProgress picks the signal. It scans for progress bars with their current percentage,
percentage readouts in text, spinners and busy regions, each with a selector. If it finds nothing,
the page shows nothing measurable, and you are asked what completion looks like instead of being
given a watch on nothing.
Limits
| Concurrent monitors | 3 |
| Default duration | 30 minutes |
| Maximum duration | 4 hours |
| While it runs | The tab is pinned; you can work anywhere else |
| On completion | A card on the page you are on, plus the run's own report. While the browser is behind other windows or minimized, the Browsentic app shows it as your computer's own notification instead; clicking it brings the tab back (see Where a finished notice shows) |
Sampling is debounced and rate-limited, with a five-second backstop tick so a page that stops changing still gets checked.
Where a finished notice shows
| The browser | The Browsentic app | Where it shows |
|---|---|---|
| In front | Any | A card on the page you are on |
| Behind other windows, or minimized | Running, with notifications allowed | Your computer's own notification, from the app. Click it to bring the tab back |
| Behind, or minimized | Not running, or notifications off | A card on the page, or the browser's notification when no page can show one |
The same goes for a scheduled task finishing and for a reminder. On a Mac, allow notifications for Browsentic in System Settings › Notifications; on Windows, in Settings › System › Notifications.
Stopping one
stop monitoring
This is an instant command: it ends the watch without starting an agent. The tab is unpinned and no notification is shown, since you asked for the stop.
From an MCP client
Over the optional MCP endpoint, page_awaitMonitor long-polls a monitor until it completes. A reply
with settled: false means the poll window ended while the watch continues: call again. That is
normal, not an error. If the call fails with EXTENSION_OFFLINE, the monitor is still running in
the browser: reconnect and call again.
See also
- reference/tools.md § Monitoring: every parameter
- Scheduling: when the page shows nothing to watch and the job must be re-done
- Skills: the
monitor-progressskill routes these requests - internals/subsystems.md: how sampling works