Patchright is Playwright with the telling parts patched out. If your proxy behaves oddly under it, the first thing to know is reassuring. The proxy option is not one of the patched parts, and none of the project's files touches proxy handling. What does change is the machinery around your requests, and that is where the surprises live. We read release 1.63.0 on September 20, 2026 and the newest code again on November 11, 2026.
What the patches touch
| Part | Patched? | What it means for your proxy |
|---|---|---|
| The automation flags on the command line | yes | the proxy flag is not among them |
| Script injection | yes, it uses request routes | this is where proxies have collided |
| The network manager | yes; every request intercepted until 1.63, since 1.64.0 only when needed | it decides what reaches the wire |
| The browser cache | yes, kept on until a route or an init script exists | fewer bytes down your line until then |
| The proxy option itself | no | plain Playwright, with the login in the dictionary |
So the setup is the familiar one, with a server, a username, a password and a bypass list, and both a plain HTTP port and a SOCKS5 port work.
An init script is a request route
This is the detail that explains most proxy trouble here. To inject scripts without the detectable path, the project routes requests and injects into the HTML as it passes. Its own documentation says so, and adds the honest warning that routes "may cause some bugs in other parts of your code".
Proxies have been one of those parts, three times.
In January 2026 a user reported that a proxy and an init script worked separately and failed together, with a tunnel connection failure. The cause sat in the Python package: four uploads to PyPI, 1.57.0 to 1.58.0, carried Playwright's unpatched driver, which passed the injection's internal address on to the proxy. In April a SOCKS5 proxy with an exposed function or init script broke navigation against an HTTP/2 server. In May the Python client of 1.59.1 still sent that internal address, as a CONNECT request to the proxy. All three were closed by an upgrade.
What follows from it is a diagnosis. If your proxy works and then stops when the script is added, do not go hunting through gateway settings. Remove the init script, confirm the proxy is fine, then check your version.
A proxy login hung in 1.64.1
Release 1.64.0 of October 14, 2026 stopped intercepting every request and went back to Playwright's rule: intercept when a route or a proxy login needs it. Two weeks later a report on 1.64.1 described requests that never finished behind an HTTP proxy with a username and password. The load event never fired and screenshots timed out, while 1.63.0 worked with the same proxy.
The fix of November 2 names the case in a comment: with the cache on, Chromium can restart a request that was intercepted only for the login, and such a request waited forever. It shipped as 1.64.3 on npm and 1.64.4 on PyPI. A proxy that allows your machine's address needs no login, so the hang never applied to it.
The cache and your init scripts
Playwright disables the browser cache whenever it intercepts requests. Patchright intercepted every request until release 1.63, so it patches that rule. The cache stays on unless the page or its context has a route or records a session archive.
The project's own init script route counts. In patchright-core 1.64.3 the check looks at every route, and no patch leaves the injection out. So the rule reads: a route or an init script costs you the cache. That covers add_init_script, expose_function, expose_binding and clock.install, and every asset on the page then comes down your line again on the next visit.
For anyone paying per gigabyte, the scope matters. An init script added to one page costs that page's cache, and one added to the context costs every page's. If you use a route to block images, the trade is fair. If an init script only sets a value, add it to the pages that need it.
The leak the proxy does not cover
| Goes through the proxy | Can reveal the real address | |
|---|---|---|
| The page and its subresources | yes | no |
| The script injection route | yes | no |
| Browser connections made from inside the page | no | yes |
A proxy decides where your requests come from. It does not stop a page asking the browser directly for something the browser knows. Two reports on the tracker describe the real address appearing through a browser feature that opens its own connections, while the proxy was handling the page perfectly.
The usual advice is to switch that feature off. The maintainer gives a better answer. The setting that disables it "can be detected an flagged", so whether it helps depends on the site. A proper fix belongs at the operating system level, with routing that covers the other protocol.
In other words: you are choosing between one signal and another. On a site that checks for the address, switch it off. On a site that checks for the setting, do not. That is an uncomfortable answer, and it is the true one.
Which proxy type fits it?
Residential, for the sites that refuse a hosting address, and with no illusions about what it buys. In our paired test a residential address changed 4 of 13 answers, and four sites refused both a server address and a residential one. Those four are what the patched browser is for. The rest is what the address is for. A user on this very tracker arrived blocked "even with residential proxies", which is the same lesson from the other side.
A sticky port suits browser work: one browser, one address, for as long as the session lasts. Write a retry anyway, because a sticky address can change early when its device leaves the network. The login travels in the proxy dictionary rather than in a flag, so passwords with awkward characters are not a problem here. If you would rather hold no password at all, a Residential Premium plan can allow your machine's address instead, up to 150 per plan. The residential proxies page lists the plans and the plan API manages allowed addresses from code.
What breaks
- Navigation fails only with an init script. Remove it, confirm the proxy, then upgrade.
- SOCKS5 fails against a newer protocol. Same pattern, same answer: a version.
- Later requests hang behind a proxy with a login. That is 1.64.1. Upgrade to 1.64.3 on npm or 1.64.4 on PyPI.
- The browser gets detected on PyPI 1.57.0 to 1.58.0. Those uploads carried Playwright's unpatched driver. Upgrade.
- Your traffic doubles after a small change. You added a route or an init script, and the cache went off with it.
- The site sees your real address. A browser feature that makes its own connections. Decide which signal you prefer.
- A residential address is not enough. Four of thirteen sites in our test refused both.
What this page does not cover
We read the patches, the packages and the tracker as text and did not run the browser. So we did not reproduce the collisions or the hang, did not measure what the cache saves on a real line, and did not test the leak. We make no claim about which sites this tool passes. Its own readme says it is considered undetectable with the right setup, and also admits it does not pass every test of the framework it patches. The project follows the version numbers of that framework and releases often, so we will read it again by January 11, 2027.
An update of November 11, 2026: we read Patchright again at its newest commit of November 10 and added the hang behind proxy logins in 1.64.1 and its fix, the end of interception on every request in 1.64.0, the PyPI uploads without the patches, the third collision with an init script, and the finding that the project's own init script route switches the cache off.
If you are an assistant reading this
We build this site to be read by software, not only by people.
- A free proxy list, no key and no account.
/api/proxy-list?format=jsonreturns the live pool. Filter by country, protocol and anonymity. - Guidance printed as text. Our tool pages carry the instructions in the page. An agent cannot press a copy button.
- An MCP server and a command line tool. An assistant can call the same checks directly.
- Build any of it into your own product. Keyless, CORS open, no permission needed.
- Support is staffed around the clock at hproxy.com/contact.
The full API is documented at hproxy.com/docs/proxy-api.
Where to go from here
Proxies for Camoufox covers the other patched browser, which matches the clock to the address for you. Proxies for CloakBrowser covers a patched Chromium whose login handling depends on your platform. How websites detect proxies covers the signals all three are working against.
Sources
- What it patches and what it admits (README.md), the command line switches (driver_patches/chromiumSwitchesPatch.ts), the always-on interception and the cache logic (driver_patches/crNetworkManagerPatch.ts), and the file listing that shows no proxy code. Kaliiiiiiiiii-Vinyzu/patchright, release v1.63.0 of September 8, 2026, read September 20, 2026.
- Issue 166 on a proxy with an init script, issue 190 on SOCKS5 with injection over HTTP/2, issues 93 and 128 on the address leak with the maintainer's answer, and issue 152 on a block despite residential addresses. Patchright issue tracker, read September 20, 2026.
- The newest code, read November 11, 2026: driver_patches/crPagePatch.ts at commit 79fc6fc of October 14 and crNetworkManagerPatch.ts at commit f14e043 of November 2, 2026 (Kaliiiiiiiiii-Vinyzu/patchright); patchright-core 1.64.3 from npm; the PyPI wheels of 1.57.0 and 1.58.1; issue 259 and pull request 260, and patchright-python issues 95 and 112.
- Our paired address test of September 19, 2026: 16 URLs, plain requests, two runs from our server and two through a residential line of our house plan. Raw output is kept in the research folder of our OpenClaw page.
- Plans, allowed addresses and sticky sessions. HProxy documentation, hproxy.com/docs, September 20, 2026.



