Download websites that use Zipkin

Finding a distributed tracing backend from the public internet is unusual, and the fact that it can be found at all is most of what a record here means.

Zipkin came out of Twitter in 2012, modeled on Google's Dapper paper, released under the Apache 2.0 license and written mainly in Java. A tracing system collects timing data as a request passes through a chain of services, then reassembles the fragments into one trace so an engineer can see where the latency went. The design assumes microservices. Nobody installs tracing for a brochure site.

The population is therefore self-selecting in a useful way. Organizations that appear in this list run service-oriented architectures, employ enough engineers to need shared observability, and have already decided to spend attention on it. For vendors selling monitoring, error tracking, log management, service meshes or platform engineering tools, such a domain is a qualified prospect rather than a guess, and a team running an open-source tracer is a team with a build-or-buy conversation already open.

The caveat is large and should be read before the number is. Tracing infrastructure normally lives inside a private network, so the sites where we can see it are the ones that exposed a user interface, a collector endpoint or an instrumented page to the public web. As of October 2026 we record 12,415 of them, a small and skewed slice of real usage: public demos, documentation deployments, staging hosts left open, and genuine misconfigurations. The invisible part of the installed base is far larger than the visible part.

One more limit is worth naming. Several tracing stacks speak a compatible protocol, and a project that has moved on to OpenTelemetry may still expose endpoints that look like Zipkin from the outside. A match is best read as compatible tracing rather than as proof of the project itself. For a security researcher the ambiguity hardly matters, because an exposed tracing surface deserves a look either way. For a commercial researcher the list still does what a lookup cannot, identifying a population of engineering organizations by their architecture rather than by their industry code, with the rest of each host's detected stack sitting in the same row.