After Being Banned by Claude N Times, I Finally Set Up Full Overseas Infrastructure (Beginner Guide)

I have been banned by Claude X times, and through that process, I've summarized a lot of experience. Now, I finally have stable access to Claude.

Here, I will detail the entire configuration process (timezone, IP, proxy settings, phone number, payment card) so you never get stuck with Claude bans again!!!

First, thanks to several experts whose experiences are integrated into this guide:

@app_sail, @leo_xiaolei, @suyuan1711, @JANEX0609, @Nemo001ai.

In this article, I document the complete process. Some steps might be critical, others might be unnecessary, but since no one knows exactly why Anthropic bans accounts, I am sharing everything I did and learned from other users!!!

▌Most Important Insight: Bans Are Multi-Factorial

Claude rarely bans you for a single reason. It aggregates signals like timezone, IP, proxy config, phone number, payment card, and usage behavior. Once these accumulate past a threshold, your account is gone.

This explains why some people use shared airport nodes for two years without issue, while others with clean environments get banned in days.

Our goal isn't to find one specific cause, but to clean up every signal so we appear as a normal, stable overseas user to their risk control system.

Let's break it down step by step.

▌Device: Clean or Replace Devices That Have Been Banned

If your device has had multiple accounts banned, the device itself may be flagged. New accounts registered on it might start with a negative tag.

My advice involves two steps.

Step 1: Local cleanup. Claude leaves traces on your computer, mainly here:

1. The .claude hidden folder in the user root directory.
2. Claude credentials stored in macOS Keychain. Many people miss this; deleting just the .claude folder isn't enough.
3. Historical session files in Application Support. If you know the org_id of the banned account, search for it and delete related files.

If you don't want to do this manually, give this prompt to your Agent: "Help me clean up local Claude remnants on this computer, including the .claude directory, Claude credentials in Keychain, and historical session files in Application Support. List them for confirmation before cleaning."

If your device has been banned, user @suyuan1711 created a Skill specifically to clear device fingerprint records. He saved two or three banned MacBooks with it. I haven't tried it, but if your device is problematic, it's worth a shot:

https://github.com/suyuan2022/suyuan-skill/blob/main/claude-cleanup/SKILL.md

If none of this works, consider replacing the device...

▌Timezone: Set Phone and Computer to Taipei

Someone decompiled the Claude Code client and found it does check timezones.

Taipei is within Claude's allowed range and shares the same UTC offset as Beijing Time, so there's zero impact on daily use. Just change it.

Mac:

1. System Settings → General → Date & Time. Turn off "Set time zone automatically" and manually select "Taipei".
2. Go to Privacy & Security → Location Services → System Services and turn off "Set Time Zone".

iPhone:

1. Settings → General → Date & Time. Turn off "Set Automatically" and select Taipei.
2. Settings → Privacy & Security → Location Services → System Services, turn off "Set Time Zone".

▌Phone Number: A Real US Number

Registration and risk control check the phone number's origin. SMS verification platforms can pass registration but are easily flagged later. Strongly recommend getting a long-term personal US number.

I wrote a separate full tutorial on this. Red Pocket costs $30/year and works with BeeSIM on domestic phones. The whole process from ordering to writing the SIM is covered:

https://x.com/gkxspace/status/2101984042635944146

Just follow that guide; I won't repeat it here.

▌Network: The Most Critical Factor

Network and payment cards are the biggest factors in bans. Let's discuss network first.

Do not use shared "airport" nodes. If you share an IP with thousands of people, and even one person is doing distillation or account farming, the risk score for that entire IP segment rises.

I offer two routes. Choose based on your situation.

Route 1: Self-hosted VPS (For those willing to tinker)

Buy a US VPS and set up your own node. The entire chain is used only by you—clean, controllable, and cheap.

Choose providers like DMIT or BandwagonHost, which are well-verified. For extra stability, you can layer residential IPs behind the VPS, though this is optional.

Node setup tutorial after purchase (give this to your AI): https://x.com/gengdaJ/status/2036803153841328388

Routing rule configuration tutorial: https://x.com/gkxspace/status/2094386340645216468

Route 2: Easy Setup with IPEqual

If you don't want to manage servers, use IPEqual directly.

It has the highest purity residential IPs I've seen, combining nodes and US residential IPs in one package. Buy once, get everything, no self-setup required.

I recommend the Max membership. The price is similar to a standard VPS, and if your usage isn't extremely high, this is sufficient.

🔗: https://www.equaldcdn.com

Three Must-Do Configurations After Choosing a Line

The line is just the foundation. Configuration details often cause failures.

1. Use TUN mode, not system proxy. Standard system proxy modes leave records like 127.0.0.1:7890 in macOS network settings. Claude Code detects this. TUN mode creates a virtual network interface at the system level, keeping system proxy settings empty. To Claude Code, it looks like a regular computer connected directly to the US internet.

Verify in terminal: Run scutil --proxy. HTTPEnable and HTTPSEnable should be 0. Then run curl ipinfo.io to confirm the exit IP is in the US. Both must be correct.

1. Clean up historical proxy configurations. Check shell environment variables, .zshrc / .bashrc, npm and git proxy configs, ~/.claude/settings.json, and all anthropic-related environment variables.

Give this to your Agent: "Comprehensively check this computer for proxy remnants, including system proxies, proxy env vars, shell startup files, npm/git proxy configs, claude settings.json, and all ANTHROPIC-prefixed env vars. List anything unclean."

1. If the line drops, let Claude disconnect. Never auto-fallback to direct connection. This is crucial. If you usually use Claude via proxy, and one day the node fails, the proxy client might switch to direct connection. Claude sees your IP instantly jump from the US back to China.

Correct approach: Assign a specific route for Claude traffic. If that route fails, reject the connection. Do not allow automatic node switching or direct fallback. Check your routing rules to ensure claude.ai and api.anthropic.com cannot fall to DIRECT. Ensure the default rule is not direct connection.

Also important: Phone and computer must use the same verified line (if you use mobile).

Use this command for final checks (from @app_sail):

If any output is not empty, not 0, or not null, manually clean it.

1# 1. System Proxy (should all be 0)
2scutil --proxy | grep -E "HTTPEnable|HTTPSEnable|SOCKSEnable"
3
4# 2. Current shell environment variables
5env | grep -iE "proxy"
6
7# 3. Hardcoded proxy in shell startup files
8grep -iE "proxy" ~/.zshrc ~/.bashrc ~/.profile ~/.zshenv 2>/dev/null
9
10# 4. npm / yarn / pnpm configs
11npm config get proxy
12npm config get https-proxy
13cat ~/.npmrc 2>/dev/null | grep -i proxy
14
15# 5. Git configs
16git config --global --get http.proxy
17git config --global --get https.proxy
18
19# 6. Claude Code's own settings.json
20cat ~/.claude/settings.json 2>/dev/null | grep -iE "proxy|base_url|anthropic_"
21
22# 7. All anthropic / claude related env vars (focus on ANTHROPIC_BASE_URL)
23env | grep -iE "anthropic|claude"
24
25# 8. Homebrew
26brew config | grep -i proxy

▌Email: Register a Fresh Gmail

I simply registered a new, clean Gmail and used it to sign up for Claude. After signing up, I subscribed to Pro first. Do not use purchased pre-made emails. If you have an unbanned Gmail, you can use that too.

▌Payment: Another Critical Factor

Domestic dual-currency credit cards and various virtual cards either fail to pay or land you on a risk control list immediately.

1. Best case: You have a personal US bank card, or you establish a US offshore company to apply for a US card. This is the most normal payment method to risk control systems.

1. US PayPal also works. Link a domestic multi-currency card or HK card... then link PayPal to a US Apple ID. This also works.

1. If you don't have the above, use Starryblu. It's an excellent Singapore physical card that I currently use. It has the lowest barrier to entry among all options.

3.1. Card opening link: https://sg.starryblu.com/x/6RtMh9c0. Or download the Starryblu APP, use invite code: MQFK01Q, and fill out the whitelist form: https://docs.qq.com/form/page/DRmhsUEFHbUZhYkx5

3.2. Wait 1-2 days. After the card is issued, buy the $9.9 plan (or preferably $19.9 for more benefits). This covers shipping, and then you can use it permanently.

3.3. This is a Singapore card. When purchasing Claude membership, remember to switch your node to a Singapore IP. You only need to do this during the initial purchase. Subsequent renewals are automatic, and you don't need to worry about IP changes.

https://x.com/gkxspace/status/2089668722105688198

Downside: Subscribing to Claude with a Singapore card uses Singapore pricing. For Max 20x, it costs about 100 RMB more per month than normal. But if you lack other payment methods, paying a bit extra for a stable renewal channel without monthly anxiety is worth it.

1. Gift cards. However, if you get banned, refunds seem unavailable.

1. More solutions summarized by users: https://x.com/Nemo001ai/status/2072882143336767816

▌Usage Behavior: Don't Be Too Aggressive Initially

1. Don't start with max specs. I signed up, got Pro, upgraded to Max 5x after three days, and Max 20x after ten days. (Direct Max 20 might be fine, but being cautious is better...)

1. Users who max out 20x are the ones Anthropic loses money on. Try not to go full throttle immediately...

1. Don't repeatedly ask it to do things it has explicitly refused. Claude logs behavioral patterns.

The principle is simple: Make Anthropic believe this account is a normal person using the service cleanly.

▌Reflection

That's roughly my summary after tinkering for half a year. It includes the full suite of overseas environment and essential infrastructure.

With this complete configuration, you should be able to use Claude stably. Moreover, this setup is nearly zero-barrier, considering various potential scenarios.

Follow me for more high-quality content soon~

If you encounter other issues, post them in the comments, and I'll help answer!!!