Claude for Chrome
Claude Code is already installed and already signed in. Browsentic spawns it against the tab you are looking at, so the plan that writes your code now reads and drives the page. No second subscription, and no API key to paste.
Chrome, Brave and Arc
Claude
for
Chrome.
Gemini
for
Chrome.
Codex
for
Chrome.
One
extension.
Browsentic is a Chrome extension with an AI side panel, and it brings no model of its own. It hands the tab in front of you to the agent CLI already sitting on your machine, inside the session you are already signed in to.
Pick the agent
Whichever one you already pay for
The popup lists every agent CLI it found on your PATH. Switching is one click, and it changes who is reasoning, not what the panel can do.
Gemini for Chrome
Antigravity is Google’s agent CLI and Gemini is the model behind it. Pick it in the popup and the side panel is Gemini, working in your own Chrome profile rather than a clean one it launched for itself.
Codex for Chrome
OpenAI’s Codex CLI, on the ChatGPT plan you already have. Same panel, same tools, same tab. Only the reasoning changes, and you can change it back mid afternoon.
Mistral Vibe
betaMistral’s CLI, wired up the same way. It is in beta here because its session handling is still settling down, so expect the odd retry on a long run.
Grok Build
betaxAI’s CLI. Also beta, for the same reason: it works, it has had less mileage in the panel than Claude Code and Codex have.
What Chrome gives you
The browser with the debugger gets the full set
Chrome exposes the DevTools Protocol, and nine of the tools are built on it. This is the build where none of them are missing.
All 41 tools, nothing held back
The trusted click, both captcha tools, the four diagnostics tools and the two page-code tools all need Chrome’s debugger. On Chromium they are there. Firefox has no equivalent, so its build offers a shorter list.
Chrome says when a debugger is attached
You get the "Browsentic is debugging this browser" bar for as long as one is held: about 250 ms for a trusted click, until you stop it for a diagnostics recording. Attaching fails while DevTools is open on that tab, because Chrome allows one debugger per tab.
A real side panel
Not a floating window and not an overlay injected into the page. It is Chrome’s own side panel, docked beside the tab, and it stays put as the page navigates underneath it.
Brave and Arc take the same build
One Manifest V3 extension covers every Chromium browser. Load the same folder at brave://extensions or in Arc and pair it the same way. Edge has its own page, because Copilot makes the comparison worth spelling out.
Install
Two minutes, and two of them are yours
One command does the machine half. The two steps inside the browser are the ones only you can take, by design: nothing reaches your browser without you putting it there.
-
01
Run the command
The npm package carries the extension build, so nothing is cloned and nothing is compiled. It writes the extension out, starts the local daemon and prints a pairing code.
shnpx browsentic setup -
02
Load it at chrome://extensions
Turn on Developer mode, press Load unpacked, and choose the folder the command printed. On macOS press Shift Command G in the picker and paste the path; on Linux, Ctrl+L. Pin Browsentic so the popup is one click away.
path~/browsentic/extension/chrome-mv3 -
03
Paste the pairing code
It is single use and lives ten minutes. Paste it into the popup, press Connect, and the daemon hands back a session key that survives restarts. Run browsentic pair for a fresh one.
shnpx browsentic pair
On a Mac the app does all of this from a window, Node included: curl -fsSL https://browsentic.com/install.sh | sh
Questions
What people ask about Chrome
Is there a Claude extension for Chrome?
Anthropic ships one, called Claude in Chrome, and it runs Anthropic’s agent on a paid Claude plan. Browsentic is the other shape: an MIT licensed extension that runs whichever agent CLI you already have, Claude Code included, from a daemon on your own machine. There is a full comparison on its own page.
Can I use Gemini in Chrome through this?
Yes, by way of Antigravity, Google’s agent CLI, which runs Gemini locally. Install it, log in once, and pick it in the Browsentic popup. That is separate from Google’s own Gemini in Chrome feature, and the two do not get in each other’s way.
Does it work in Brave, Arc and other Chromium browsers?
Yes. It is a Manifest V3 extension, so anything Chromium loads it: Brave, Arc, Vivaldi, Opera and Edge. Load the same unpacked folder from that browser’s extensions page and pair it there.
Why is it loaded unpacked instead of from the Chrome Web Store?
Because the extension and the local daemon ship as one version and have to stay in step. The npm package carries the build, so npx browsentic update refreshes both at once and your pairing survives. Press the reload arrow on the Browsentic card afterwards, since Chrome never reloads an unpacked extension on its own.
Can it see the sites I am logged in to?
It acts in the tab you point it at, with the cookies and the two-factor state that tab already has. That is the point: no throwaway profile, no signing in again. Nothing is read until you ask for something in that tab, and anything that reaches another person stops for your approval first.
Stop describing the page. Hand it over.
Free and MIT licensed. No signup, no key to paste, and a fresh install connects to nothing until you pair it yourself.
npx browsentic setup