Selenium, Playwright and Puppeteer are great tools. They were built for one job above all: driving a browser from the outside, usually headless, usually in CI. When you need to test a web app at scale, that is exactly what you want.
But a lot of browser automation is not testing. It is the boring, repetitive work people do every day in their own browser: copying values from one back-office into another, opening the same five monitoring pages every morning, filling the same form for the tenth time, checking a status on a supplier's portal. For that kind of work, the classic stack feels wrong:
- a separate WebDriver process and often a separate browser profile, without the user's sessions;
- a script the user cannot see, running somewhere they do not control;
- often a server or a cloud service in the middle.
I wanted the opposite: automation that runs in the user's real browser, in tabs they can see and stop, only on sites they explicitly approved, with nothing leaving the machine. That is what Remote Tab Opener (RTO) does. This post explains the model and shows the minimal code.
The model: a trusted page, a local extension, an allow-list
There are three pieces:
- Your own web page. It can be served from
localhostor anyhttps://origin you control. This is where your workflow lives: buttons, logic, UI. - The RTO browser extension (Chrome or Firefox). It receives requests from your page through
window.postMessage, checks them, and acts on real tabs. - An explicit allow-list. The extension is deny-by-default. A page cannot authorize itself: the user approves each host on a "master" tab, and the extension refuses everything else.
Your page talks to the extension, the extension talks to the tabs. No server, no proxy, no relay. The tabs are normal tabs in the user's window, with the user's existing sessions, and the user can watch, take over or close them at any time. An optional overlay cursor even shows where the automation is about to click.
I wrote more about why visibility matters here: Visible browser tab automation: what it means in practice.
Minimal code
With the extension installed, include the SDK Lite bundle (RTO.min.js, MIT-licensed, on GitHub) and detect the extension first:
<script src="RTO.min.js"></script> <script> RTOconnectOnce(2000).then(function (ok) { if (!ok) return console.warn("Remote Tab Opener extension not detected"); console.log("RTO connected"); }); </script> Then ask the user to approve the host you want to work on. This is not a silent auto-allow: the request shows up on the master tab and the user decides.
RTOAllowlist.add("example.com", 130000).then(function (ok) { console.log(ok ? "Approved" : "Cancelled by the user"); }); Now open, or reuse, one controlled tab. The tabKey is a stable name: calling openTab again with the same key reuses the same tab instead of piling up new ones.
RTOForm.openTab( { tabKey: "demo", url: "https://example.com", focus: true, newTab: false }, { timeoutMs: 20000 } ).then(function (res) { console.log("Controlled tab:", res && res.tabId); }); From there you can run a small flow: open a page, wait for an element, read some values. Every message uses the same public bridge vocabulary:
sendRto({ type: "flow.run", tabKey: "demo-flow", steps: [ { type: "openTab", args: { url: "https://example.com", tabKey: "demo-flow" } }, { type: "dom.waitFor", args: { selector: "h1", timeoutMs: 15000 } }, { type: "dom.extract", args: { fields: [ { name: "heading", selector: "h1", mode: "text" } ] } } ] }).then(function (response) { if (response.ok) console.log(response.data); else console.error(response.error); }); (sendRto is a ten-line postMessage helper with a request id and a timeout; it is given in full in the AI-safe quickstart.)
The SDK Lite repository has three runnable examples: a minimal connection page, a visible cursor walkthrough and a complete local demo against a test form.
What RTO does not do
The constraints are the point, so let me be explicit:
- It does not bypass the Same-Origin Policy. Your page never gets direct access to another site's DOM; the extension performs documented actions and returns results.
- It does not act on sites the user has not approved. No allow-list entry, no action.
- It is not a scraping farm. One user, their own browser, their own sessions, visible tabs. If you need hundreds of headless browsers in parallel, use Playwright.
- It does not send your data anywhere. No server, no cloud relay: requests go from your page to the extension and to the tab, all inside the browser.
Website owners who do not want to be automated at all can opt out. The full picture is in the security model.
When to use what
| You need… | Use |
|---|---|
| End-to-end tests in CI, headless, at scale | Playwright / Selenium / Puppeteer |
| A one-off script you run yourself | A userscript or the browser console |
| A shared internal tool that drives real tabs in the user's browser, visibly, on approved sites | RTO |
A longer comparison is here: RTO vs Selenium, Playwright and Puppeteer.
Try it
- Extension: Chrome · Firefox (free)
- SDK Lite and examples: github.com/CodeVersonLab/rto-sdk-lite
- Docs: remote-tab-opener.com/documentation
I built RTO, so I am obviously biased. I would really like feedback on the security model and on the use cases you would try it for. Comments are open.

