A SOCKS5 proxy is configured, pages open, and an IP-check site shows the proxy's address. That looks like a working setup, but it only covers TCP traffic. Calls, WebRTC, QUIC and HTTP/3 run over UDP, and UDP can take a different route from the one the page uses.
To test a proxy for UDP leaks, one question matters: does UDP traffic leave from the same address as TCP traffic? The sections below describe six public checks that answer the question from different angles, and the limits of the result.
Six checks, six angles
No single page covers every UDP-based technology, so six public tools are used together. Each one looks at a different protocol or channel.
| # | Tool | What it tests | A pass looks like |
|---|---|---|---|
| 1 | Twilio Network Test | WebRTC calls over UDP | All checks pass, including TURN UDP connectivity, and the test calls finish with no packet loss |
| 2 | BrowserLeaks QUIC | QUIC support and the address seen over QUIC | QUIC connection is true and the IP matches the proxy exit address |
| 3 | WebRTC Detection Check | Addresses over HTTP, STUN, TURN UDP and TURN TCP, plus spoofing flags | All four sources show the same IP and no detection flag is raised |
| 4 | NGINX QUIC Sandbox | HTTP/3 under sustained load, 3,000 requests | Every request succeeds over QUIC, with no fallback to HTTP and no failures |
| 5 | wtcheck.top | WebTransport address against the TCP address | Both addresses are identical |
| 6 | BrowserLeaks DNS | Which resolvers answer DNS queries | Resolvers belong to the exit node's provider, not to the local ISP |
The reference value for every check is the same: the proxy's exit address, as shown by a regular TCP IP check.
1. Twilio Network Test: do real calls work over UDP?
This is Twilio's own diagnostic for checking whether a network can handle browser-based voice calls. It knows nothing about leaks or proxies. It simply tries to make calls, which makes it a useful neutral first step.
The key line is TURN UDP connectivity: it means UDP packets left the browser, reached Twilio's servers and came back. The tool then places two test calls with different audio codecs (PCMU and Opus) and reports packet loss and a quality rating for each. Twilio also notes that its voice service does not work over TCP, so a successful call can only have gone over UDP.
A pass is a green connectivity result and two completed calls with no packet loss. If the UDP connectivity line fails, the proxy either blocks UDP or does not support it.
September 10, 12:04. One tab—the session has just started.
2. BrowserLeaks QUIC: does UDP work for ordinary web traffic?
Check 1 covered UDP used for calls. QUIC is a different use of UDP: it carries regular web traffic and is the transport under HTTP/3, which large sites already use. Support for one does not imply support for the other, so both need a test.
The page answers two questions at once. First, whether a QUIC connection was established at all. Second, which IP address the server saw during that connection. Compare that address with the reference value from the regular TCP check.
A pass is a QUIC connection reported as true and an address equal to the proxy exit address. A different address means QUIC traffic is bypassing the proxy. A false result means QUIC is blocked.
September 10, 12:04. There are now two tabs—Twilio is still open on the left.
3. WebRTC Detection Check: the same address across four channels
This page compares how the browser's address looks through four routes: a plain HTTP request over TCP, a STUN request over UDP, a TURN relay over UDP and the same relay over TCP. If the proxy handles UDP correctly, all four show one address.
It also raises detection flags for suspicious setups, such as WebRTC blocked, a mismatch between UDP and TCP addresses, spoofed ICE candidates, blocked TURN traffic or a fake relay. The flags matter because some setups hide the real address by switching WebRTC off or by feeding it substitute values. Both approaches are visible here: disabling WebRTC trips the first flag, and substitution trips the spoofing flags.
A pass is four identical addresses and no raised flag. A mismatch between the TCP line and the STUN line is the classic leak.
September 10, 12:05. Third tab, with the two previous tests still open on the left.
4. NGINX QUIC Sandbox: does HTTP/3 hold up under load?
Check 2 opens one QUIC connection. This test sends 3,000 requests in a row and sorts them into three counters: succeeded over QUIC, succeeded over plain HTTP, and failed. It shows whether the channel stays stable, not only whether it starts.
The second counter is the most informative. Browsers fall back to regular HTTP when QUIC does not work, so a page that loads normally proves nothing about QUIC. A fallback count above zero means some requests could not use UDP and quietly took the TCP route.
A pass is all 3,000 requests over QUIC, zero over HTTP and zero failed. Many fallbacks or failures point to an unstable UDP path, which matters for long sessions.
September 10, 12:07. Fourth tab.
5. wtcheck.top: WebTransport as a third UDP path
WebTransport is another browser feature built on QUIC. It is tested separately because each protocol can behave differently, and a passed WebRTC or QUIC test does not rule out a problem in a third path.
The page makes two requests and shows both results side by side: the address of an ordinary TCP request and the address of the WebTransport connection.
A pass is two identical addresses. Two different addresses mean WebTransport traffic leaves outside the proxy.
September 10, 12:10. Fifth tab.
6. BrowserLeaks DNS: who answers the name lookups?
DNS is a separate channel. A proxy can be configured correctly while the browser still asks the local provider's resolver for site addresses. This has two effects: the provider can see which sites are visited, and a website can notice that the resolver's location does not match the IP address's location.
The test lists the DNS servers that took part in resolving its queries, together with the provider that owns each one. The useful figure is the number of different providers. In a clean setup, the resolvers belong to the exit node's provider.
A pass is resolvers owned by the same provider as the exit address, or by a public DNS service chosen on purpose. The red flag is the local ISP's resolvers appearing in the list. A resolver from a different company is not automatically a leak, so the owner and location matter more than the raw count.
September 10, 12:11. Sixth tab, with all previous tests still open on the left.
How to run the checks and read the results
The checks give more information when they run in one continuous session, so that nothing changes between them.
- Open a browser or profile with the SOCKS5 proxy configured, and note the exit address from a regular IP-check site. This is the reference value.
- Open the six pages from the table in separate tabs one by one, without closing earlier tabs and without changing proxy or browser settings in between.
- For each page, compare the address it reports with the reference value.
- Check the DNS result last, and look at who owns the resolvers rather than at a single IP.
The failures follow a few patterns:
- A different address in checks 2, 3 or 5. UDP is leaving outside the proxy, so a site can see the real address.
- QUIC is false, WebRTC is reported as blocked, or the Twilio test fails on UDP. UDP is blocked or the proxy does not support it. Many SOCKS5 proxies are TCP-only, so this is worth confirming with the provider.
- Resolvers from a different provider in check 6. DNS queries bypass the proxy. The local resolver can then see the browsing history, and a site can compare the resolver's location with the IP address.
- Check 4 falls back to HTTP or reports failed requests. UDP works but not reliably under load, which matters for long sessions.
What these checks do not prove
A passing run shows that one proxy behaved correctly in one browser at one moment. It does not show that the setup is safe in general.
- The result belongs to a single combination of browser version, proxy and exit address. A different proxy, a rotating exit IP or a browser update can change it.
- Public test pages change or go offline, and their detection logic is not a standard. A page that reports no flags is not a guarantee against every anti-fraud system.
- Some test services limit requests per IP, so a shared proxy address can produce an error unrelated to leaks.
- The tests cover the channels listed in the table. Other traffic, such as extensions or system-level requests, is outside them.
For anything that matters, the checks should be repeated after changing the proxy, the browser version or the profile settings, and a rotating proxy should be tested more than once.
Example run in Octo Browser
Octo Browser, an anti-detect browser that keeps a separate profile (fingerprint, proxy, cookies) per task, supports UDP over SOCKS5. WebRTC, QUIC, HTTP/3 and WebTransport traffic is routed through the proxy instead of bypassing it, and the profile bar shows the UDP exit address next to the proxy, which gives a ready reference value for the checks above.
The six checks were run in an Octo Browser profile with a SOCKS5 proxy, and none of them showed a leak: all six reported the same exit address for TCP and UDP. This is a single run, so the same procedure should be repeated with your own proxy before relying on it.
Conclusion
To test a proxy for UDP leaks, compare six angles against one reference address: WebRTC calls, QUIC, address matching across four channels, sustained HTTP/3, WebTransport and DNS ownership. Passing all six describes one proxy at one moment, so the run is worth repeating whenever the proxy or browser changes. To try the profile-based setup from the example, the promo code DEVTO gives new users 4 days of the Starter plan.






Top comments (0)