chrome-sidecar: Let AI Agents Use Your Logged-In Browser

TL;DR chrome-sidecar lets an AI agent take over the Chrome you already use — reusing your login sessions, working in background tabs, and pausing at logins, CAPTCHAs and payments.

You’ve probably run into this.

You ask an AI to look something up on some website, and it says, “I’m ready — please log in first.” A browser window pops up. You type your username, wait for the SMS code, get past the picture-picking CAPTCHA, and when you’re finally done you go back and tell it “okay, done.” Switch to another site, and you do it all over again.

I got tired of that flow, so I built something. The idea is simple: don’t give the AI a brand-new browser — let it use the one you’re already using.

Why it always turns into “log in first”

Most current approaches to letting AI drive a web page give the AI its own separate browser.

Separate browsers have their advantages — they’re clean and don’t touch what you’re doing. But the cost is that the AI knows nothing. It has no access to your accounts or to anything you’ve already set up. So the first step is inevitably: log in.

Some approaches copy your browser profile for the AI to use, which at least brings the login state along. But your login state only lives in one copy, so either you use that copied browser forever, or the two copies slowly drift apart. Swapping out my main browser just to run some automation doesn’t seem like a good trade to me.

A different idea: sit the AI next to you

The thing I built takes another path.

It doesn’t open a new browser and it doesn’t copy the profile. It takes over the Chrome you’re already using. Sites you’ve logged into are logged in for it too; your extensions and your tweaked preferences are all there.

It’s a bit like having a colleague sit at your desk, use the pages you already have open to get something done, and hand you the result when they’re finished.

The difference is that it follows a few rules:

It works in the background and never takes over your screen. New pages open in background tabs — the tab you’re viewing never gets yanked away, and your typing is never interrupted. You can let it work while you get on with your own things.
When a step genuinely requires you, it stops and waits. Logins, CAPTCHAs, QR scans, payment confirmations — it never touches any of these. It just tells you “this step needs you,” then waits, clicking nothing in the meantime. When you’re done, you say so and it carries on.
It never touches your passwords. It won’t type them for you, won’t read your verification codes for you, and won’t go read what’s in the password fields on your page.

What you can use it for

These are roughly the things I use it for day to day:

Open my dashboard and tell me yesterday’s revenue
Pull the content off some page and drop it into my notes
Fill in a form in some system, upload a screenshot, and submit it
Check whether an order has shipped, and if so post the tracking number in the right thread
Sync one piece of content to several platforms
Chores like changing account settings that require me in person — it opens the page, and I handle the rest

If it’s something you’d do yourself by clicking around a web page, and it needs a login first, it’s a good candidate.

How to install it

Two steps, set up once.

Step one: flip a switch in the Chrome you normally use. Type chrome://inspect/#remote-debugging in the address bar and tick “Allow remote debugging for this browser instance.”

This is an official Chrome switch that lets external programs connect to the browser. When something connects, Chrome pops up a dialog asking “Allow remote debugging?” — just click allow.

That dialog is a security boundary, and I deliberately did not make the tool auto-click it. When you see it, please click it yourself.

Step two: install the skill into your AI tool.

It’s an instruction manual plus tools for the AI — install it into the AI tool you normally use (Claude Code, Cursor, Codex, WorkBuddy, and so on) and it’s ready.

I recommend managing it with flint, which distributes the skill to all your AI tools so one change updates them everywhere:

1 2	git clone https://github.com/lcy362/chrome-sidecar npx flint-skills-hub

If you’d rather not use a tool, copying the chrome-sidecar/skills/chrome-sidecar folder into your AI tool’s skills directory works just as well.

How to use it once installed

There’s nothing to start and no commands to memorize. Just talk to your AI:

Open my dashboard and tell me yesterday’s revenue.
Screenshot that page and put it in my notes.
Check whether this order has shipped.
Help me change my account settings — call me when a login is needed.

That’s it. The rest is its job.

A few prerequisites

It needs Chrome or a similar browser, and it has to be the one you actually use. Opening a new browser just for it won’t work.

The switch above has to be on; I can’t bypass it for you, because that’s the browser’s own security design.

And one honest note: if all you want is for the AI to read a public web page, don’t use this — just let it read the page. This thing is built for “tasks that require you to be logged in.”