Playwright connect_over_cdp does not work with Firefox

You reached for browser_type.connect_over_cdp(...), pointed it at a Firefox instance, and got nothing to connect to. Most pages that answer this bury the reason under a workaround. The reason is the whole answer: connecting to Firefox over the DevTools Protocol is not hard, it is structurally impossible, because Firefox ships no DevTools Protocol surface for anything to attach to.

This page is why the method is Chromium-only, what actually carries automation to Firefox instead, the connect() path that replaces connect_over_cdp, and why with this project you usually do not need either.

connect_over_cdp is a Chromium-only entry point

connect_over_cdp does exactly one thing: it speaks the Chrome DevTools Protocol to an already-running browser that is listening on a CDP endpoint. Playwright’s own reference for the method states plainly that “connecting over the Chrome DevTools Protocol is only supported for Chromium-based browsers.” It is a real method on every browser type in Playwright’s Python API, including firefox, so calling it does not raise “no such attribute”. It raises because there is no endpoint on the other end.

from playwright.sync_api import sync_playwright

with sync_playwright() as p:
    # Chromium: attaches to a browser already listening with --remote-debugging-port
    browser = p.chromium.connect_over_cdp("http://127.0.0.1:9222")
    # Firefox: there is no CDP port to point this at, so there is nothing to attach to

The Chromium line works because Chromium can be told to open a DevTools endpoint. The Firefox line has no equivalent to point at. The DevTools Protocol is a shipped, first class part of Chromium; it is not part of Firefox at all.

Why there is no DevTools endpoint to connect to

Playwright does not drive Firefox through the DevTools Protocol the way it drives Chromium. It drives it through Juggler, a separate automation protocol built for Firefox, carried over a pipe rather than a CDP socket. There is no CDP layer sitting underneath waiting to be exposed, so no flag turns one on.

This is not a gap that a future release closes. It is a design difference between the two engines, and for automation it cuts in Firefox’s favour: a protocol that was never shipped as a public browser interface is also a protocol with no shipped artefacts for a detector to look for. The structural version of that argument is here.

Juggler is not a loose, best-effort protocol either. Its command schema is validated closed-world: a payload carrying any field the schema was not told to expect is rejected outright, at the moment the call is made. That strictness is invisible until a client and server disagree, at which point a routine Playwright upgrade can take out 97 of 133 end-to-end tests while the browser still launches and loads pages. The point for this page is narrower: the wire Firefox listens on is Juggler, not CDP, so the connection primitive that reaches it is not connect_over_cdp.

The path that replaces it: connect() over a Juggler endpoint

The remote-connection method for Firefox is connect(), not connect_over_cdp(). It attaches to a browser server that publishes a ws:// endpoint speaking Juggler; Playwright’s own reference describes connect() as attaching “to an existing browser instance created via BrowserType.launchServer” and does not restrict it to Chromium the way connect_over_cdp is restricted.

You start that server with Playwright’s own tooling. The launchServer() call in the Node API, or the playwright launch-server --browser firefox CLI it backs, both boot a Firefox and print a ws:// URL. That URL is a Juggler websocket endpoint, not a CDP one, and you hand it to connect() from any client, Python included:

from playwright.sync_api import sync_playwright

# WS is printed by a browser server, e.g.
#   playwright launch-server --browser firefox
# It is a Juggler ws:// endpoint, not an http:// CDP endpoint.
WS = "ws://127.0.0.1:5000/<guid>"

with sync_playwright() as p:
    browser = p.firefox.connect(WS)      # NOT connect_over_cdp
    page = browser.new_page()
    page.goto("https://example.com")
    print(page.title())
    browser.close()

The mapping to remember, if you are porting Chromium code:

Chromium Firefox
chromium.connect_over_cdp(http_url) firefox.connect(ws_url)
--remote-debugging-port opens a CDP endpoint launch-server publishes a Juggler ws endpoint
DevTools Protocol over a socket Juggler over a pipe / websocket

There is no line in that table where a CDP URL reaches Firefox, because no such URL exists.

With invisible_playwright you rarely need connect at all

With invisible_playwright you usually need neither connect() nor connect_over_cdp: the launcher hands you a real, fully configured Browser object directly, in-process, and it is a genuine Playwright Browser with every standard method. People reach for connect_over_cdp because they assume attaching to a separately launched browser is the only way to get one configured the way they want. With this project it is not.

from invisible_playwright import InvisiblePlaywright

with InvisiblePlaywright(seed=42) as browser:
    page = browser.new_page()
    page.goto("https://example.com")
    print(page.title())          # every standard Playwright method works unchanged

The browser returned is a playwright.sync_api.Browser, so new_page, new_context, page.goto, page.click and the rest behave exactly as documented upstream. No CDP endpoint, no server to boot, no websocket to wire up. The seed makes the whole thing reproducible: seed=42 yields the same GPU, canvas hash, audio context, fonts and screen on every run, which is the difference between replaying a failure and guessing at it.

The async surface is identical:

from invisible_playwright.async_api import InvisiblePlaywright

async with InvisiblePlaywright(seed=42) as browser:
    page = await browser.new_page()
    await page.goto("https://example.com")
    print(await page.title())

If you genuinely need an out-of-process browser to attach to (a long-lived server, a different host), the connect() path in the previous section is the one to use. For the common case, the context manager removes the reason people went looking for connect_over_cdp in the first place.

What “just launch it normally” buys you over attaching

Attaching over a protocol has a cost that the in-process launch avoids: the connection method itself is part of what a page can measure, not just what it lets you do.

An automation protocol is a surface. Attaching a debugger to a page realm, historically, did four separately observable things to Firefox: it disabled the optimising JIT so the page ran measurably slower, it dropped the driver’s own frames into Error.stack, it labelled evaluated code as debugger evaluation, and it warmed a serialisation helper in the page’s own realm before the page’s first script ran. All four are written up, and all four are fixed here. The lesson that carries over is that how you connect is not cosmetic: the connection path is part of what a page can measure. A launcher that owns the whole engine can close those; a thin attach-over-a-port model inherits whatever the endpoint exposes.

So the honest framing is not “connect_over_cdp is missing a feature”. It is that the CDP attach model does not exist for Firefox, the Juggler connect() model does, and the in-process launch is the one that lets the engine be patched underneath you rather than observed through a socket.

Conclusion

connect_over_cdp does not work with Firefox because Firefox ships no DevTools Protocol for it to reach. Playwright drives Firefox through Juggler, a closed-world protocol over a pipe, and the remote-connection method that matches it is connect() to a ws:// endpoint from a browser server, not connect_over_cdp() to a CDP URL. With invisible_playwright the shortest path is neither: the launcher returns a real Playwright Browser in-process, seed-reproducible, with the full API intact. Reach for connect() only when you truly need a browser in another process.

Short answers to the questions that lead here

Does connect_over_cdp work with Firefox? No. It speaks the Chrome DevTools Protocol, and Firefox exposes no CDP endpoint for it to attach to. The method exists on the Firefox type but has nothing to connect to.

Why is connect_over_cdp Chromium-only? Because the DevTools Protocol is a shipped part of Chromium and is not part of Firefox at all. Playwright drives Firefox through Juggler instead.

How do I attach to a running Firefox with Playwright? Start a browser server (playwright launch-server --browser firefox, or launchServer() in the Node API), take the ws:// endpoint it prints, and pass it to firefox.connect(ws_url), not connect_over_cdp.

What is Juggler? Firefox’s own automation protocol, carried over a pipe, that Playwright uses in place of the DevTools Protocol. Its schema rejects any undeclared field at runtime, which is why a client and server that disagree fail sharply.

Do I need connect_over_cdp to configure the browser? No. With invisible_playwright the launcher returns a fully configured, real Playwright Browser in-process, so there is no separate endpoint to attach to for the common case.

Is there any CDP flag I can enable in Firefox to make this work? No. There is no CDP layer underneath to switch on. Use the Juggler connect() path instead.

Sources

  • Playwright’s own BrowserType API reference, where connect_over_cdp and connect are documented as distinct methods, the former stated as Chromium-only.
  • This project’s Juggler protocol contract, including the closed-world schema validation and the driver-layer artefacts documented on the pages linked above.

See also: Firefox or Chromium for anti-detect automation for why no-CDP is a structural property rather than a missing feature, why a Playwright upgrade broke 97 of 133 tests for how strict the Juggler contract is, and the two-line switch from plain Playwright for the in-process launch.

Written while maintaining invisible_playwright, a Firefox patched at the C++ level driven by stock Playwright. connect_over_cdp is the question we get most from people porting Chromium code, and the answer is that Firefox never had the endpoint to begin with.

Back to top

MIT licensed. Every page here is written against the current source of the thing it describes, and several record something we got wrong first.

This site uses Just the Docs, a documentation theme for Jekyll.