Skip to content
Glossary

What is a transparent proxy?

Definition

A transparent proxy usually means one the network routes traffic through without any client setting, so the client does not know it is there.

Three different things are documented, configured and sold under this name, so before answering a question about one, find out which is meant.

Meaning Where you meet it What “transparent” refers to Clearer name
Traffic is redirected to the proxy by the network, with no client setting Network operators’ caches and gateways, Squid’s intercept and tproxy modes, the Linux kernel’s “Transparent proxy support” The client cannot see the proxy in its own configuration Interception proxy (RFC 3040)
The proxy forwards messages without modifying them beyond what proxying requires RFC 2616, the old HTTP/1.1 specification The messages pass through unchanged Non-transforming proxy (RFC 9110)
The proxy passes your real address to the destination in a header Free proxy lists and online proxy checkers Your address is visible to the site Say which headers it adds

RFC 3040 records the collision between the first two. It notes that the caching community used “transparent proxy” for zero-configuration deployments, that this clashed with the HTTP definition, and introduces “interception proxy” for proxies that receive traffic redirected by network elements. It recommends always qualifying the older term. RFC 9110 has since dropped the word altogether and talks about transforming and non-transforming proxies instead. The third meaning has no specification behind it at all.

How interception works

A router, firewall or gateway sends the client’s packets to the proxy instead of the destination, by policy routing, by NAT redirection, or on Linux with TPROXY, which delivers them without rewriting the destination address. The client still believes it is talking to the origin. Squid documents both styles on its http_port directive: intercept for “IP-Layer NAT interception” and tproxy for Linux TPROXY, where outgoing connections can even use the client’s own address. Both disable proxy authentication on that port, because the client never knew to send credentials.

Plain HTTP carries the target host in every request, so the proxy can serve it without the client’s cooperation. HTTPS is different. Without TLS interception, which needs a root certificate the client already trusts, the proxy can pass the connection through, log the destination, or refuse it, but it cannot read it. A deployment without a trusted root can therefore handle plain HTTP fully, and can only treat port 443 as an opaque flow.

Checking your own connection

Where interception covers port 80 but not 443, comparing the same echo over HTTP and HTTPS shows it. A header on the HTTP answer that the HTTPS one lacks points at something in between:

curl -s http://httpbin.org/headers
curl -s https://httpbin.org/headers

From our test machine on 21 October 2026 the two answers matched apart from the trace ID, and api.ipify.org reported the same address over both, which is what an unintercepted line looks like. Then we sent the HTTP request through a local Squid 6.13 to prove the test can come back positive. It nearly did not: httpbin.org’s front end strips Via and X-Forwarded-For, so the only trace of Squid was Cache-Control: max-age=259200, a header curl never sends. A detection method is only as good as the echo server behind it, and a quiet answer proves little, because an interception proxy can be configured to add nothing.

The proxy-list meaning, measured

The third meaning turns out to be a single configuration line. With Squid’s defaults, the server behind it received Via: 1.1 squid-test (squid/6.13) and X-Forwarded-For: 127.0.0.1: a “transparent” proxy in proxy-list terms, because the client’s address was disclosed. With forwarded_for delete and via off, the same Squid sent neither header and would be sold as “elite”. What the header carries, and why its first entry cannot be trusted, is on the X-Forwarded-For entry. So “transparent” in a proxy list describes how one server was set up, not a kind of proxy, and the only way to know what an endpoint does is to check the exit yourself.

How it differs from the proxies you configure

A forward proxy does the same job with the client’s consent: you name it in the application or in HTTP_PROXY and HTTPS_PROXY, and the client sends it absolute URLs or CONNECT requests, as described in what actually happens on the wire. An interception proxy receives ordinary requests that were addressed to someone else. A reverse proxy is deployed by the server’s operator in front of the server, not by the client’s network.

Frequently asked questions

Does transparent mean the proxy cannot be detected?
No. It means the client was not configured to use it. An interception proxy can give itself away through Via entries, cache headers or its error pages, but one set to add nothing can be hard to spot, so a clean test proves little.
Can a transparent proxy read my HTTPS traffic?
Not by redirection alone. Reading it requires terminating TLS with a certificate your client trusts, which means a root the operator installed on your device. Without that it sees the destination and the volume, not the content.
Is a transparent proxy the same as a forward proxy?
The mechanics are close, but a forward proxy is one the client is configured to use, so the client sends it proxy-style requests. An interception proxy receives requests the client addressed to the origin.

Sources

  1. RFC 3040: web replication and caching taxonomy, defining interception proxy
  2. RFC 2616: HTTP/1.1, the original definition of a transparent proxy
  3. RFC 9110: transforming and non-transforming proxies
  4. Squid configuration: http_port intercept and tproxy modes
  5. Linux kernel documentation: Transparent proxy support (TPROXY)

All terms, A to Z