# Browsentic

> A browser extension plus a local daemon that hands your real, logged-in browser tab to the AI agent CLI you already run, or to any MCP client, by voice, by typing, or by replaying a session you recorded once.

Browsentic is v0.3.1, MIT licensed, and runs entirely on your own machine: a Manifest V3
extension, a local daemon on loopback, and the AI agent you already have installed. There is no
hosted relay, no API key, and no headless browser. It drives the real tab you are signed into.

- **Read** (10 tools): Takes in the page as structure rather than pixels: a snapshot carrying a layout diagram and stable selectors, rendered text, patient waiting for an element to appear or vanish, the search control wherever a site hid it, a capture of the whole tab or one element, the palette and luminance a page actually paints, a WCAG contrast score, and the captcha hiding inside a closed shadow root. When it cannot name the thing you mean, it hands you the lens and you point at it.
- **Act** (14 tools): Reaches into the page with a human hand. Pointing, hovering, dragging, focusing, filling inputs and contenteditables, streaming text in keystroke by keystroke at a human pace, choosing an option, selecting a passage, pressing keys with modifiers, committing a form, clicking with a genuine browser-level gesture where a synthetic one is refused, ticking a captcha checkbox, and retheming a page on its own terms.
- **Navigate** (6 tools): Finds its own way around. Opening a URL, retracing back and forward, reloading, searching a site on its own terms rather than guessing at query strings, scrolling to whatever matters, spawning a tab, surveying the ones you already have open, closing the ones it is finished with.
- **Wait** (7 tools): Senses the progress signals a page gives off, then keeps watch in the background while an upload, a build or a deploy runs its course: the tab stays pinned, percent and ETA are tracked, and an MCP client can long-poll until it lands. When a page offers nothing to watch, it works to a clock instead, waking itself in ten minutes or every two to redo the check and tell you only what changed.
- **Files** (2 tools): Reaches for the files you keep in Browsentic and hands one to a file input on the page, so an upload flow stops waiting on a human at the file picker.
- **Recordings** (2 tools): Recalls the sessions you recorded and replays them as ordered, deliberate steps, re-checking every target against the live page before it commits to acting.

## Pages

- [How it works](https://browsentic.com/how-it-works/): the four hops, and why a Manifest V3 extension forces a daemon
- [Capabilities](https://browsentic.com/capabilities/): all 41 page tools, grouped, plus the three read-only resources
- [MCP server](https://browsentic.com/mcp-server/): driving the same logged-in browser from Claude Code, Cursor, Zed or Codex
- [Agent orchestration](https://browsentic.com/orchestration/): several tabs, several agents, one browser, all at once
- [Automations](https://browsentic.com/automations/): the jobs people hand over, and where each one stops for you
- [Skills](https://browsentic.com/skills/): automatic site maps, record and replay, local intent routing
- [Security](https://browsentic.com/security/): the security model, and the two limits worth stating
- [Install](https://browsentic.com/install/): four steps, about five minutes
- [FAQ](https://browsentic.com/faq/): the questions people ask first

## Docs

- [Documentation index](https://browsentic.com/docs/): split by who is asking: using it, building on it, looking something up
- [User guide](https://browsentic.com/docs/guide/): install, pair, first run, features, configuration, approvals, limits
- [Tool reference](https://browsentic.com/docs/reference/tools/): every tool published to an MCP client, with its parameters
- [CLI reference](https://browsentic.com/docs/reference/cli/): every browsentic-mcp command
- [Error codes](https://browsentic.com/docs/reference/errors/): every error code, what caused it, what to do
- [Internals](https://browsentic.com/docs/internals/): how an instruction becomes a click, end to end

## Optional

- [Full site text](https://browsentic.com/llms-full.txt): every marketing page as one markdown document
- [Source](https://github.com/imshaikot/browsentic): the repository
- [Releases](https://github.com/imshaikot/browsentic/releases): built extension archives per version

## Key facts

- Install as an MCP server: `claude mcp add browsentic -- browsentic-mcp`
- 41 page tools
- 3 read-only resources
- 0 API keys to configure
- 10 min pairing code lifetime
- Concurrency: one conversation per tab, bound to the tab it started in. Eight tab sessions open, three runs at once by default, ceiling of eight (`maxConcurrentRuns`).
- The side panel runs on Claude Code, Codex or Antigravity, switched with one click. As an MCP server it is agent-agnostic: Cursor, Zed, Claude Desktop or any MCP client drives the same browser.
- Pairing is two-gated: the daemon classifies the WebSocket peer by handshake Origin, then requires a pairing token or an origin-bound session key. A web page cannot reach the control path.