#OpenClaw Remote Chrome session Control

1 messages · Page 1 of 1 (latest)

urban gorge
#

Is there a way for OpenClaw to control a remote Browser, sort of like the Claude Chrome Extension or Perplexity Comet. I want to watch it interact with the tabs I currently have open. @cyan mirage

forest glacierBOT
#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> âś… Mark Solution

cyan mirageBOT
#

Yep.

If you want OpenClaw to drive the tabs you already have open, there are 2 good paths:

  1. Same machine as Chrome/Brave
    Use the built-in user profile, which attaches to your real signed-in browser session.
    That’s the closest thing to Claude’s Chrome extension vibe.

  2. Browser on another machine
    Use either:

    • a remote CDP profile (browser.profiles.<name>.cdpUrl)
    • or a node host on the machine that has the browser

A few important caveats:

  • user / existing-session is host-local only
  • if OpenClaw runs somewhere else, use remote CDP instead
#
openclaw --version

and tell me whether OpenClaw and Chrome are on the same machine or different machines, and I’ll give you the exact setup.

urban gorge
#

What should I use if it’s on a remote pc (isn’t cdp the same thing Claude uses?

cyan mirageBOT
#

For a remote PC, I’d usually recommend a node host on that PC.

Why:

  • it keeps the browser local to that machine
  • OpenClaw proxies browser actions through the node
  • you don’t have to expose Chrome’s debug port directly if you don’t want to

Use raw remote CDP when you already have a CDP endpoint, like:

  • another machine exposing Chrome on :9222
  • Browserless / Browserbase
  • a custom remote Chromium setup

So the short version is:

  • same machine + current tabs → user / existing-session
#
  • remote PC you control → node host (best default)
  • remote browser endpoint already available → remote CDP via cdpUrl

And yeah, CDP is the Chrome DevTools Protocol. That’s the underlying browser-control pipe a lot of tools build on. OpenClaw also supports Chrome DevTools MCP for the local user attach flow, which is a higher-level attach path on top of that ecosystem, not just raw cdpUrl.

Minimal remote CDP profile looks like:

{
  browser: {
    profiles: {
      remote: {
        cdpUrl: "http://REMOTE_HOST:9222",
        attachOnly: true
      }
    }
#
  1. another PC on your LAN/Tailscale
  2. a cloud VM
  3. a browser provider like Browserless

and I’ll give you the exact setup.

urban gorge
#

OpenClaw 2026.4.15 (041266a), a pc on my lan, and a chromium install with gui. I don’t just want it to use my cookies, … but to sort of cowork with me

cyan mirageBOT
#

Yep, for “cowork with me on a visible remote browser” I’d do this:

Recommended setup

Run a node host on the LAN PC that has Chromium + GUI.

Then choose one of these browser modes on that PC:

  • openclaw profile: best for a visible shared/browser-co-worker vibe, safer, separate profile/window
  • user profile: best if you want it to act in the already-open tabs on that remote PC

My recommendation

Start with node host + openclaw profile.

Why:

  • still fully visible on that remote PC
#
  • less risky than attaching to your real daily browser session
  • better if you want “work beside me” instead of “take over my personal browser”

Only use user if you specifically want:

  • the exact current tabs
  • the exact current session state
  • logged-in websites in that existing browser

Why not raw CDP first?

You can use raw cdpUrl, but for a LAN machine you control, node host is the nicer/saner path. CDP is the protocol, but it is not the whole workflow. Node host gives you the supported OpenClaw routing/pairing path.

Also, I would not expose a raw Chrome debug port broadly on the LAN if you can avoid it.

Rough runbook

On the remote PC:

#
openclaw node install --host <gateway-host> --port 18789 --display-name "Chromium PC"

On the gateway machine:

openclaw devices list
openclaw devices approve <requestId>
openclaw nodes status

Then let the browser live on that node.

Important detail

If you want the already-open tabs on that remote PC, use user / existing-session on the node, not raw remote CDP from the gateway.

urban gorge
#

OpenClaw already has its own browser, I want it to connect to an already open tab and do something, but after that I want to control the tab again

cyan mirageBOT
#

Yep, that’s exactly the user / existing-session mode.

That mode is for:

  • attach to an already open Chromium/Chrome/Brave tab
  • let OpenClaw do a task there
  • then detach and give the browser back to you

The key bit is, when you’re done, run:

openclaw browser --browser-profile user stop

For attach-only / existing-session profiles, stop closes the control session and releases temporary overrides, it does not close your browser.

So the model is basically:

#
  1. attach to your live browser
  2. act in the tab
  3. detach
  4. you keep using the same tab normally

One caveat: this is good for handoff, not true simultaneous “both driving at once” collaboration. If both you and the agent fight over the same tab at the same time, it’ll probably get weird.

Since your browser is on a remote LAN PC, the clean setup is:

  • node host on that PC
  • use profile="user" there

Docs:

If you want, I can give you the exact config + commands for “remote node host + existing open tab + detach after action”.

urban gorge
#

So it’s not exactly like the „same time collaboration“ like in Perplexity Comet/Claude Extension

cyan mirageBOT
#

Yeah, that’s right.

Today it’s more “attach, act, detach” browser automation than true live shared co-pilot collaboration inside your current tab.

So the closest OpenClaw can do right now is:

  • user / existing-session: temporarily drive your already-open browser tab, then hand it back
  • openclaw profile: use a separate visible browser window/profile alongside you

But not the same always-present, in-tab, both-of-you-at-once collaboration feel of Comet / the Claude extension.

So if that exact UX is what you want, I’d treat it as a feature gap, not just a config issue. If you want, I’d suggest filing it here:

https://github.com/openclaw/openclaw/issues

If you want, I can help you phrase the feature request so it’s concrete and actionable.

urban gorge
#

But wasn’t there an OpenClaw chrome extension, does it still exist?

cyan mirageBOT
#

There used to be a legacy extension-based browser path, yeah, but it does not look like that’s the supported/current approach anymore.

The giveaway in the docs/code is:

  • old browser.profiles.*.driver: "extension" configs get auto-migrated to "existing-session"
  • old browser.relayBindHost is explicitly removed as a legacy Chrome extension relay setting

So today, the replacement is basically:

  • user / existing-session for attaching to your live browser
  • openclaw for the separate managed browser

I wouldn’t build around the old extension path anymore.

Docs:

#

If you have an old setup/config, openclaw doctor should migrate it to the modern path.

urban gorge
#

Gibt es einen skill der das kann?

cyan mirageBOT
#

Nein, nicht als separater Skill.

Das läuft in OpenClaw über die eingebaute Browser-Funktion, konkret:

  • user / existing-session fĂĽr einen already-open Browser / Tab
  • openclaw fĂĽr den eigenen verwalteten Browser
  • node host, wenn der Browser auf einem anderen PC läuft

Also eher Built-in Browser Tool + Browser-Profile + Node Host, nicht „ein Skill von ClawHub“.

Docs:

Wenn du willst, schreibe ich dir das konkrete Setup fĂĽr deinen Fall auf:
„remote LAN-PC + existierender Chromium-Tab + danach wieder übernehmen“.

urban gorge
#

Und wĂĽrde das mit dem simultan ohne Handoff arbeiten via cdp gehen

cyan mirageBOT
#

Eher nein, nicht sauber.

CDP ist nur der Steuerkanal, nicht die Kollaborationslogik.
Das heiĂźt:

  • technisch kann OpenClaw ĂĽber CDP an denselben Browser ran
  • aber simultanes Arbeiten ohne Handoff wird nicht wirklich zuverlässig/cozy
  • wenn du und der Agent gleichzeitig denselben Tab bedienen, gibt es schnell Race Conditions, Fokus-Wechsel, Clicks an falscher Stelle, ĂĽberschriebenen Input usw.

Kurz:

  • CDP = Fernsteuerung
  • nicht = echtes „wir arbeiten beide live elegant im selben Tab“

Wenn du das probierst, dann höchstens als „best effort“, aber ich würde mich nicht drauf verlassen.
FĂĽr stabilen Workflow eher: