Run a WebRTC leak test
This WebRTC leak test answers a specific question: did a readable public address candidate differ from this browser’s HTTP reference address? The summary gives the finding, its meaning and a next step. Read the individual observations before deciding whether a difference violates your intended VPN or proxy configuration.
Read more
Connect to the network you want to inspect, then select Run WebRTC leak test. Keep Wi-Fi, VPN settings and the selected server unchanged during collection. Switching connections halfway through can mix observations from different routes and times. Allow up to twelve seconds; Stop ends collection and preserves any observations already received as incomplete evidence.
WebRTC candidate gathering contacts Cloudflare’s public STUN endpoint and separate ipify HTTP endpoints for IPv4 and IPv6. Those providers receive the source address of each request. No camera or microphone permission is needed. The tool creates no call with another person and does not upload the candidate report to our application server. A service’s own request processing is separate from the report kept in this tab.
Start with the result summary, then inspect the HTTP reference cards and candidate details. Copying is optional and includes observed addresses. Review the report before sharing it, particularly local addresses. Clear removes the displayed observations and stops an active run; it does not remove a report you already copied to another application.
Understand the three possible findings
Different public address observed means a readable, non-relay public candidate or public related address differed from an available HTTP reference of the same address family. This is a reason to investigate routing. It does not identify an address as your residential ISP or prove that your VPN failed. Another interface, an expected exit or an intentional split-tunnel policy can explain a difference.
Read more
No difference observed means candidate gathering completed without a reported STUN error and all compared public observations matched their references. The conclusion covers the available candidates and these endpoints in this run. It cannot certify DNS routing, applications outside the browser or protection after a future network change.
Inconclusive means the evidence does not support either conclusion. Examples include a missing same-family reference, no readable public candidate, an unfinished collection or an error affecting the run. A blank list is never treated as a privacy pass. Execution status describes collection; the business conclusion describes what can be inferred from its observations.
IPv4 and IPv6 are compared separately. An IPv4 value and an IPv6 value are different representations of different address families; comparing their strings would produce a meaningless warning. If the IPv6 reference fails, an IPv6 candidate remains uncomparable. A valid IPv4 observation is retained rather than discarded because another request failed.
What the candidate details describe
A host candidate describes an interface made visible by the browser. A server-reflexive, or srflx, candidate describes an address learned through STUN. The parser also distinguishes peer-reflexive and relay candidates. This test configures STUN rather than a TURN relay or a remote peer; a relay address is excluded from the unexpected-address comparison because it belongs to a relay service.
Some browsers substitute an mDNS name ending in .local for an interface’s literal address. That is a concealed address observation, not an IP we can compare. The tool does not resolve the name or guess a hidden value. A private address such as 192.168.1.20 describes a local network context and is different from a public internet exit.
Candidate rows identify type, transport, scope and any readable related address. Unspecified placeholders such as 0.0.0.0 or :: are not presented as discovered addresses. The saved report omits the complete session description and temporary connection credentials. These choices keep the details useful without implying that every browser-reported field describes a unique person or device.
Investigate differences and incomplete runs
Repeat the test without changing connections. Check whether the same difference recurs and whether both values belong to the same family. Note the browser, selected VPN location, time and intended policy. A repeatable observation with an expected address is more useful to support than a screenshot labelled “leak” without context.
Read more
Review the protection you actually configured. A browser extension, system VPN and application-specific proxy can cover different traffic. A split-tunnel exclusion may be deliberate. Verify the expected exit with your provider and, when helpful, inspect registration or routing information through IP lookup and ASN lookup. A database country difference alone does not establish unwanted exposure.
A timeout means collection did not report completion within our limit. It does not prove that WebRTC is disabled or that protection worked. An ICE error concerns an attempted endpoint or path; another attempt may still produce a candidate. Preserve the error and any successful observations together. If the peer-connection API is unavailable, use a supported browser and rerun rather than interpreting the unavailable state as a match.
This page does not observe your recursive DNS resolver or test every application on the device. Use DNS leak testing for resolver observations when its service is available, and IPv6 testing for protocol-specific HTTP evidence. Neither result replaces a documented routing policy. Our browser privacy test matrix provides dated examples of why gathering state and privacy conclusions need separate interpretation.