Why Claude was the relay killer app
Anthropic supports fewer countries than OpenAI and enforces it at signup, billing and API level. For developers outside the list, relays like AIProxy were the path of least resistance: someone else's Anthropic account, resold by the token. The convenience was real — and so was the single point of failure, as every former AIProxy balance-holder now knows.
The direct setup
- A static residential IP in a supported country (US is the safest default). This is the foundation: Anthropic sees a normal home connection at signup and on every API call.
- An Anthropic account with region-consistent details and a payment method that matches. Virtual cards issued for the account's region are the standard answer.
- Optionally, a gateway. One API / New API / LiteLLM translate between OpenAI-format requests and Anthropic's Messages API — so every OpenAI-compatible tool you own can talk to Claude through one endpoint.
Consistency: the rule that keeps accounts alive
Every lost Claude account we've seen post-mortemed shares one trait: origin inconsistency. Registered on one VPN, used from another country, billed from a third. The risk systems don't need to detect the proxy — they just flag the impossibility. The fix is boring: one static IP, used for everything, forever. Registration, dashboard logins, API traffic, billing updates — all through the same address.
- Pin the IP in your gateway config so API traffic can't accidentally bypass it.
- Keep a browser profile with matching timezone/locale for dashboard visits.
- Never reuse that IP for scraping or other accounts — it's infrastructure, not a utility pool.
Relay vs direct for Claude, side by side
| Factor | Relay | Direct + static residential |
|---|---|---|
| New Claude models | When the relay adds them | Launch day |
| Long-context & beta features | Usually missing | Full access |
| Prompt privacy | Relay logs everything | Only Anthropic sees data |
| Balance risk | Total (see AIProxy) | None — your account |