CONNECT-IP: the boundary between proxies and IP tunnels

Core idea

CONNECT-IP carries IP packets over HTTP and therefore operates below an HTTP request proxy. Applications may send multiple protocols through the tunnel, but only within routes and policy negotiated with the gateway.

For CONNECT-IP: the boundary between proxies and IP tunnels, distinguish device, router, proxy endpoint, exit IP, ASN, and geolocation estimate. They describe different points in a network. A database may return a regional center or carrier gateway, so an IP result must never be presented as a person's exact physical address.

What the primary source establishes

Compare request-level proxying with an IP tunnel and verify routes, allowed prefixes, and protection of internal networks.

The primary source, RFC 9484 — CONNECT-IP, RFC 9298 — CONNECT-UDP, RFC 9297 — Capsules, RFC 9000 — QUIC, RFC 8200 — IPv6, defines the technical baseline but not every client and provider configuration. Read the normative behavior with its version and then verify your implementation. Treat anything beyond the source as a product feature that needs separate confirmation.

Controlled lab

Create client, target, and service networks in isolated namespaces. Advertise only the target documentation prefix. Capture routes before and after setup, reach the permitted echo service, reject the service address, close the session, and verify that tunnel state disappears without changing the client's normal default route.

Step-by-step verification

Build two isolated documentation networks, allow one prefix, and deny the other. Record routes before setup, test an owned endpoint and return path, then verify route removal when the session closes.

Draw the client-to-endpoint path and mark DNS, NAT, and proxying. Capture direct and proxied address family, exit IP, ASN, claimed region, and time. Repeat mobile-network runs because carrier gateways may change without physical movement.

Compare product location, actual ASN, and at least two independent databases. City disagreement can fit normal accuracy limits. Verify country before a strict-country session; use device location with explicit permission when a workflow truly needs coordinates.

• Capture initial routes
• Allow one test prefix
• Deny the service network
• Close and verify cleanup

Evidence to retain

Record negotiated prefixes, before/after routes, interface, MTU, permitted and denied results, and teardown time. Do not publish real internal addressing.

Record a masked IP or internal ID, address family, ASN, organization, network type, database, and update date. Collect evidence in one time window. Do not publish connection credentials or unnecessary coordinates.

Define report columns and time format before the run. A result without context becomes a guess: the address, cache state, and changed condition are unknown. Record controlled failures as well as successes so the check proves that it can distinguish states.

Interpreting the result

Judge success by routing state, prefix boundaries, denial of the service network, and clean teardown rather than one open web page. An unexpectedly broad route is a security failure even when latency looks good.

CONNECT-IP is not a complete VPN product. Authentication, DNS, kill-switch behavior, and key management remain separate design decisions.

One successful run confirms only one client, route, and moment. Repeat while changing one variable and state the limits. When observation conflicts with documentation, rule out cache, client version, and intermediaries before creating a reproducible support case.

Worked decision process

Draw CONNECT-IP: the boundary between proxies and IP tunnels as device, gateway, public NAT, proxy entrance, proxy exit, DNS, and origin, with address family and measurement point. This prevents local addresses from entering remote allowlists and separates entry-server location from exit geography.

Correlate geography with ASN and observation time. Databases age differently, carrier gateways are shared, and anycast changes paths. A city mismatch is not proof of failure; verify country and exit IP in the same session as the application request.

Troubleshoot local DNS and routes, entry connectivity, authentication, exit family, and destination reachability in order. Ping does not test HTTP CONNECT. After a fix, repeat the original small scenario with the same DNS and destination.

Common mistakes

Do not allowlist 192.168.x.x on an external service, confuse CGNAT with a proxy, or treat an ISP label as proof of a mobile device. IPv6 and DNS handling also vary by client, so validate the full path with the actual application.

Stop when errors rise, a source returns a limit, the task would require bypassing protection, or secrets enter logs. Save sanitized diagnostics and correct the cause first. More concurrency or another IP can hide the fault and add load without improving evidence.

Rollout and maintenance criteria

Define the decision boundary before rollout: which observation permits continuation, which requires review, and which stops the workflow. Record acceptable error ratio, maximum wait, and the owner of every exception so a temporary failure cannot silently become permanent configuration.

Review real load, cost, and quality after the first week. Schedule a small control after client, proxy-service, or network changes. Archive outdated instructions with their replacement date and reason so operators do not follow conflicting configurations.

Operational checklist

Turn the successful experiment into a short procedure covering owner, safe configuration, limits, and stop conditions. Every run needs a terminal status. After browser, library, or network changes, run a small control before the main queue.

• Every route address is labeled
• DNS and NAT are separate
• Direct and proxy exits are compared
• Database date is known
• City is not treated as exact
• Client compatibility is verified