Find out in a few minutes whether Chrome is really using your proxy, and fix the login problem that most often breaks it.
Cause 1: the login is written into the proxy address. Chrome documents that it does not use credentials embedded in the proxy settings, so --proxy-server=http://LOGIN:PASSWORD@HOST:PORT does not log in. In a normal window you may see a sign-in box and type the login by hand; in headless there is no window to type into, so the page just fails.
Cause 2: the headed and headless runs do not use the same proxy source. The Chromium network stack uses the system network settings by default, and a desktop profile may also carry a proxy extension. A container or CI runner has neither, so the same script exits with the machine's own IP, or fails to connect.
Cause 3: protocol or port does not match. HTTP and SOCKS5 use different ports on the Proxy302 gateway (2222 and 3333), and Chrome supports no authentication for SOCKS5. If you run a login-protected SOCKS5 proxy in Chrome through Selenium, it will not work whatever the headless setting is.
A fourth thing looks like a drop: on a rotating line each new connection gets a new IP, so a long script can show a different exit IP half way through. That is how the line works, not a fault.
| Item | Login handled? | Notes |
|---|---|---|
| --proxy-server=http://HOST:PORT | No | Fine for a proxy that needs no login; otherwise the request is refused |
| Login inside the address (http://LOGIN:PASSWORD@HOST:PORT) | No | Chrome does not use credentials embedded in proxy settings |
| Selenium Proxy object (options.proxy) | No | Sets the address and type; it has no login for Chrome to use |
| Local forwarder (for example gost) that adds the login | Yes | Chrome connects to 127.0.0.1 without a login; the forwarder logs in to the proxy |
| selenium-wire | Yes, documented | The repository was archived on 3 January 2024 and is no longer maintained, so avoid it in new projects |
Read from the Chromium proxy notes, the Selenium driver options page and the selenium-wire repository on 13 November 2026. "Yes" means the documentation describes it; test with your own Chrome and chromedriver versions.
Open My assets, copy HOST, PORT, USERNAME and PASSWORD, and run the curl line in the code block below. It should print the exit IP and country. If curl fails, the problem is the proxy details, not Selenium or headless.
Expected result curl prints the exit IP of the proxy.
Chrome supports no login for SOCKS5, so give it an HTTP asset (port 2222 on Proxy302). A per-IP static lease also answers over HTTP on 2222 with the same login when it was bought as SOCKS5 (tested 4 November 2026); other assets keep the protocol they were created with.
Run the gost line below. It listens on 127.0.0.1:8080 and logs in to the proxy for you. Then start Chrome with --proxy-server=http://127.0.0.1:8080 and no login at all. Check the options of your gost version before you rely on it.
Before continuing Bind it to 127.0.0.1, not to all addresses, otherwise other machines on your network can use your proxy login.
Run the Python test below. A page that loads proves nothing, because Chrome may have gone out directly. Only the printed IP shows the real route: it should match the curl result from step 1.
Expected result The printed IP equals the one from curl.
# 1) test the login outside the browser (HTTP port)
curl -x http://LOGIN:PASSWORD@HOST:2222 https://ipinfo.io/json
# 2) local forwarder: Chrome connects here without a login
gost -L http://127.0.0.1:8080 -F http://LOGIN:PASSWORD@HOST:2222Example only: replace LOGIN, PASSWORD and HOST with the values from My assets; percent-encode special characters such as @ (%40) in the login.
from selenium import webdriver
from selenium.webdriver.chrome.options import Options
from selenium.webdriver.common.by import By
options = Options()
options.add_argument("--headless=new")
# the local forwarder from step 3: no login in this address
options.add_argument("--proxy-server=http://127.0.0.1:8080")
driver = webdriver.Chrome(options=options)
try:
driver.get("https://ipinfo.io/json")
print(driver.find_element(By.TAG_NAME, "body").text)
finally:
driver.quit()Example only. It needs Chrome and a matching chromedriver (or Selenium Manager). If it prints your own IP, Chrome is not using the proxy.
Check Where did the window run get its proxy: a system proxy, an extension, or a profile?
Fix Set the proxy explicitly with --proxy-server in the headless run, and print the IP as in step 4. Do not rely on settings that exist only on your desktop.
Check Does the curl test from step 1 pass, and is the forwarder still running?
Fix Fix the host, port and protocol first (HTTP, port 2222). The proxy connection guide lists the checks in order.
Check Is the login still written into the --proxy-server address?
Fix Remove it from the address and let the forwarder send it. Chrome ignores the embedded login.
Check Is it a rotating line? Each new connection gets a new IP.
Fix For a task that must keep one IP, use a static IP, or open a session on a dynamic line.
The gateway has an HTTP port, 2222, and a SOCKS5 port, 3333. For Selenium with Chrome use HTTP. Rotating residential gives a new IP on every new connection, which suits many independent page loads. For a test that must keep one exit, a static ISP IP does not rotate during the lease and costs US$0.50 for 1 day or US$7.99 for 30 days.
There is no free trial; the smallest purchase is one IP for one day or 1 GB of rotating residential traffic.
Not by design. Chrome now has one unified headless mode, and --proxy-server applies to it as in a normal window. What differs is where the proxy comes from and who can answer a sign-in prompt, which is why the same script behaves differently.
The Selenium docs list --headless=new among the common Chrome arguments, and Chrome's documentation says the --headless flag runs the unified mode. Either works with a current Chrome; the old mode now exists only as chrome-headless-shell.
Not with Chrome: it will not use credentials embedded in proxy settings. Use a local forwarder that adds the login, as in step 3.
Open an IP page inside the same headless session and compare the printed IP with the curl result. A page that loads is not proof.
Its repository was archived on 3 January 2024 and the page says it is no longer maintained. It may still run on old setups, but for a new project a local forwarder is simpler to keep working.
Related pages on proxies in automation.
Selenium proxy with username and password in Chrome: HTTP proxy in ChromeOptions plus Selenium's BiDi auth handler, no extensions. Static datacenter IPs for test runs.
View options and pricingThe ERR_PROXY_CONNECTION_FAILED error appears in Chrome and Edge when the browser cannot establish a connection with a proxy server.
Chrome has no SOCKS5 login, but a static ISP per-IP lease also works on HTTP port 2222 with the same login.
Set an HTTP proxy in Puppeteer with --proxy-server, page.authenticate and browser contexts, and check the exit IP.
Buy datacenter proxies by the IP: fast fixed IPs at the lowest price per address, billed per IP and term. HTTP or SOCKS5, choose the country.
View options and pricingOfficial documentation · Confirmed on:
Read on 13 November 2026: Chrome "does not implement this, and will not use any credentials embedded in the proxy settings"; proxy authentication "will instead go through the ordinary flow to find credentials"; "No authentication methods are supported for SOCKSv5 in Chrome"; with an HTTP proxy name resolution is deferred to the proxy.
Official documentation · Confirmed on:
Read on 13 November 2026: "The Chromium network stack uses the system network settings", proxy settings among them; --proxy-server takes a single URI or a scheme mapping, and --proxy-bypass-list only has an effect together with --proxy-server.
Official documentation · Confirmed on:
Read on 13 November 2026: Chrome has unified Headless and headful modes and the --headless flag runs the unified mode; since Chrome 132.0.6793.0 the old Headless mode is only available as the standalone chrome-headless-shell binary.
Official documentation · Confirmed on:
Read on 13 November 2026: commonly used Chrome args include --start-maximized, --headless=new and --user-data-dir; the Chrome and chromedriver major versions must match.
Official documentation · Confirmed on:
Read on 13 November 2026: the Python example sets options.proxy = Proxy({'proxyType': ProxyType.MANUAL, 'httpProxy': 'http.proxy:1234'}), with the Proxy and ProxyType classes imported from selenium.webdriver.common.proxy.
Official documentation · Confirmed on:
Read on 13 November 2026: "This repository was archived by the owner on Jan 3, 2024" and "Selenium Wire is no longer being maintained"; its README documents a username and password in the proxy URL for HTTP Basic Auth.
Official documentation · Confirmed on:
Read on 13 November 2026: the page shows "gost -L http://:8080 -F https://192.168.1.1:8080" style commands, where -L is a local service and every -F adds a hop of the forwarding chain. Credentials in the -F address follow the same URL form; check your gost version.
Product observation · Confirmed on:
Product facts file of the site (prices in US$, 1 PTC = US$1, protocols and ports, lease terms; protocol tests of 4 November 2026 and the rotating datacenter line of 10 November 2026). Checked on 13 November 2026.
Create an account, top up once, then pick rotating residential, static ISP or datacenter IPs and have one running in your browser within minutes.
