Split-Tunnel VPN: Which Egress IP Support Actually Sees
TL;DR: Split tunneling sends only some traffic through the VPN. Your browser’s public IP check may show the tunnel while a vendor’s API still sees your ISP—or the reverse. Capture both paths with an IP inquire lookup before you open a “please allowlist my IP” ticket.
Full tunnel versus split tunnel in one sentence
Full tunnel forces default internet egress through the VPN gateway. Split tunnel exempts routes—often local LAN, trusted SaaS CIDRs, or “everything except corporate”—so packets leave through different doors depending on destination.
Support teams hear “I’m on the VPN” and assume one egress. Split configurations make that assumption false.
Why two lookups disagree on the same laptop
A web-based IP page contacted over HTTPS follows whatever route your OS chooses for that destination’s IP. If the VPN client steers only RFC1918 and specific corporate ranges into the tunnel, a public IP website usually rides your residential ISP path. Conversely, if the client tunnels all traffic except local printers, the same website shows the VPN egress.
Browser extensions, DoH, and IPv6 preference add more forks. Always note: VPN client state, split setting, IPv4 vs IPv6 result, and whether Wi-Fi or Ethernet was active.
App-specific leaks that look like VPN bugs
- Desktop apps pinned to a physical interface
- Electron apps ignoring system proxy settings
- WSL2 or Docker bridge traffic taking a different NAT path
- Mobile hotspots while the laptop VPN is “connected”
Each can present a different public IP to the service you care about.
A practical capture sequence for tickets
- Disable VPN; record public IPv4/IPv6 and ASN.
- Enable VPN in the mode you normally use; record again from the same browser.
- From the failing application’s environment (or
curlif it is CLI), hit a simple echo endpoint and save the body. - Paste all three results into the ticket with timestamps.
If steps 2 and 3 differ, you have split-tunnel or app-path evidence—not a mystical CDN bug.
Allowlists and the wrong door
Security teams often allowlist the corporate VPN egress. Users on split tunnel who reach SaaS directly will still appear as residential IPs and get denied. The fix may be “send this app through the tunnel,” not “add my home IP.” Adding home IPs for every employee recreates the problem split tunneling tried to avoid.
Ask whether the allowlist expects the VPN SNAT address. Test by forcing that destination into the tunnel (full tunnel temporarily, or a route entry) and re-checking the public IP the SaaS admin portal logs.
IPv6 split behavior
Some clients tunnel IPv4 only and leave IPv6 native. Dual-stack websites may prefer IPv6, so your “VPN IP” page shows ISP v6 while v4 went through the gateway. Disable IPv6 temporarily for a controlled test, or record both families every time.
Writing results so future you understands them
Store: client name/version, split mode label from the UI, interface names, and the IP lookup screenshots. Avoid saying “VPN works” without the egress address. Works-for-SSH is not works-for-browser.
Reading split configuration from the client UI
Split tunnel settings are rarely labelled clearly. Some clients list “included routes” and “excluded routes,” where a short exclude list effectively tunnels almost everything and a short include list tunnels almost nothing. Confusing the two produces a test that proves the opposite of what the engineer intended.
Read the list before trusting the mode name. A setting labelled “split tunnel” with no exclusions is a full tunnel wearing a different badge, and a test performed under that misapprehension will be confidently wrong.
When the UI is ambiguous, prove the behaviour instead of interpreting the label. One route to a known destination, one lookup, one recorded result settles the question faster than any settings tour.
DNS as the forgotten leak channel
Even when web traffic rides the tunnel, name resolution may not. Queries can reach a local resolver that reveals the network you are actually on, and some filtering products treat that as your real location.
If a service misbehaves only when you are on the tunnel, and the exit address looks correct, check where DNS is going before rebuilding the client configuration. The fix is often a resolver setting rather than a routing change.
Record the resolver alongside the address in your notes. Two lines of evidence answer most follow-up questions without another round of testing.
Keeping the capture repeatable next month
Save the command sequence itself, not only its output. A short script that records the address before and after connecting turns a twenty-minute investigation into a single command the next time the question arises.
Store the results with a timestamp and the client version, since behaviour changes between releases. A note without a version is a story; a note with one is evidence.
Confirming egress with www.myipinquire.com during VPN tests
Run the same inquire page before connect, after connect, and from any container or VM that shares the host network namespace. When the three disagree, fix routing intent before chasing application bugs. A steady habit of checking www.myipinquire.com turns split-tunnel surprises into documented paths support can actually allowlist.