Proxies for Claude AI

Yes — and for a personal account it's the better setup, not just the simpler one. Set the address in the browser profile where you use claude.ai, export it as HTTPS_PROXY for the SDK, and both surfaces present one consistent identity: same country, same classification, same history. Splitting a personal account across different exits creates exactly the geographic incoherence the risk system is trained to notice.

The moment to split is when the API stops being personal: a CI pipeline hammering the API from your account's home address distorts its behavioral profile in the other direction. Teams typically keep the ISP address for humans logging in and move automated traffic to datacenter IPs — often under a separate Console account, so a flag on one never reaches the other.

Before — the order genuinely matters here. Anthropic records where an account was born, and a mismatch between registration location and everyday login location is one of the risk factors its scoring watches. An account created on your flagged office network and then moved to a clean address carries that mismatch forever; an account registered through the dedicated address it will live on has a coherent story from day one.

So the sequence is: get the address, confirm it isn't sitting on any blocklist, set it up in your browser profile, and only then sign up. Ten minutes of patience buys you a registration IP, login IP, and payment country that all point at the same place.

It's one of the higher-risk combinations, and it bites at the worst moments: signup, plan upgrades, and subscription renewals — the events where payment and network signals get compared side by side. A billing address in Germany with an exit in Singapore doesn't prove anything by itself, but it's a strong hint of account resale or a compromised card, so the automated check treats it that way.

The fix costs nothing extra: pick your proxy's location to match your billing country. If you've genuinely relocated and your card lags behind, update the payment details first, then move your exit — one signal changing at a time reads as a life event; both at once reads as a takeover.

Not the way that question usually means it, and we'd rather be straight about that. Creating a fresh account to route around a ban violates Anthropic's terms, and the new account inherits every correlatable signal — payment method, device fingerprint, usage patterns — so it tends to die faster than the first. A proxy doesn't change that math.

The path that actually works is the appeal: claude.ai/restricted or support.anthropic.com, with your account email, the date, the exact error, and a specific explanation — “my VPN rotated exit countries between sessions” gets reviewed better than a vague plea. Users do get restored, though queues run days to weeks. Where the proxy earns its place is after restoration — or on the account you haven't lost yet — as the reason the flag never fires again.

Yes — Claude Code honors the standard HTTPS_PROXY environment variable, the same convention corporate networks have used for years, so there's no special configuration to learn. Set it in the shell where you run the CLI, or in your shell profile to make it permanent, and every request to Anthropic exits through your dedicated address.

The detail worth checking is consistency with the rest of your setup. Claude Code authenticates through your Anthropic account, so its traffic contributes to the same risk profile as your claude.ai logins — a terminal exiting from one country while your browser sessions come from another recreates the split-identity pattern you bought the proxy to avoid. One address, everywhere your account touches Anthropic, is the whole discipline.