Skip to content

Capabilities

Page tools that read a page, act on it and wait out long jobs

The agent reads a page as structure: a layout diagram with stable selectors, rendered text and screenshots. It clicks, types and submits with real browser gestures, and watches long uploads, builds and deploys from the background.

Read

Read the page before acting on it

Reads the page as structure: a layout diagram with stable selectors, rendered text, screenshots, and the site’s search box wherever it sits. When the agent cannot tell which element you mean, it asks you to point at it.

Act

Click, type and submit like a person

Clicks, hovers, drags, focuses, fills, types at a human pace, chooses options, selects text, presses keys and submits forms. Where a page refuses a synthetic event, it uses a real browser gesture.

Script

Write a page script you approve

Writes a small script for the page when the fixed tools do not fit, such as a job that repeats twenty times. You read and approve the script before any of it runs. Off until you turn it on, and never available to an MCP client.

Site tools

Call the tools a site registers for agents

Where a page registers WebMCP tools for agents, the page snapshot says so, and one schema-checked call into the site’s own code replaces a click sequence that a redesign could break.

Navigate

Navigate pages and tabs

Opens a URL, goes back and forward, reloads, searches a site through its own search, scrolls to an element, and opens, switches and closes tabs.

Wait

Wait out uploads, builds and deploys

Watches an upload, a build or a deploy from the background and tracks its percent and ETA. When a page shows no progress, a timer checks it on a clock instead and reports only what changed.

Diagnose

Read console errors and failed requests

Reads console messages, uncaught exceptions, and failed requests with their timings, so the agent can tell “the button did nothing” from “the POST behind it returned a 500”.

Files

Upload and download files

Moves files both ways: puts a file you attached into a file input, and captures the file a click downloads, sealed on disk and ready to upload elsewhere.

Recordings

Replay a job you recorded

Lists and reads the jobs you recorded, so the agent can replay them as ordered steps, checking every target against the live page before it acts.

Zero tool calls

Read-only resources

Page context the agent reads directly, without spending a tool call on it.

browsentic://page/diagram

Layout diagram with stable selectors

browsentic://page/current

Structured snapshot of the active tab

browsentic://page/text

Rendered text, as a reader would see it

Targeting

How a tool targets an element

Every tool that acts on an element takes the same target fields. Prefer visible text: it survives the redesigns that break selectors. When nothing describes the element, point at it with A-Eye.

selector

A CSS selector, ideally one page_getPageInfo handed back rather than one you guessed.

text

Case-insensitive visible text the element should contain. The most durable of the four.

role

A tag name or ARIA role, to narrow the matches down.

nth

A zero-based index, for when several elements legitimately match.

Give your browser a superpower.

Free and open source under Apache 2.0, with no API key, subscription or account. One line installs Browsentic Bridge and opens the extension’s store page. Nothing connects until you pair the two.