Claude Code from mainland China: what actually blocks you
Not whether it installs, but login, network stability, and how fast it burns your quota.
Installing is the easy part. Everything awkward happens afterwards. In the order it actually bites.
1. The login step
It is a terminal tool, but authentication goes through a browser. The CLI only receives credentials once the browser flow completes. If that flow fails, the CLI keeps reporting "not signed in" — reinstalling will not help, because the problem is not the binary.
Common symptom: the browser says signed in, the CLI still returns 401. Usually the credential landed in a different user directory, or the two logins used different browser sessions. A third source is multiple machines: one login at the office, another at home, ending on two different accounts.
Debug order: sign out, close every terminal, keep one browser, confirm its account page shows the account holding the plan, then start the CLI login again.
2. Network stability
This is not request-response. It is a long continuous session: read files, run commands, read output, repeat. A flaky connection shows up as "the task stopped for no reason" or "the command hung halfway", which reads like a tool defect and is a link defect.
The real cost: every interruption rebuilds context, so the same task on a bad connection can burn several times the quota of the same task on a stable one. That, more often than a small plan, is why people feel they are out of usage.
Test it: run a ten-minute continuous task and watch whether it survives. If it does not, fix the link before worrying about quota — upgrading a plan to absorb interruptions is paying for instability.
3. Quota burn rate
The most common misjudgement is estimating agent usage from chat usage. One agent run that reads twenty files, executes five commands and re-reads the errors is not the same order of magnitude as a conversation.
The mechanism is two-layer: a rolling window of roughly five hours, plus a weekly ceiling on top. Short, intense bursts hit the five-hour window first; steady daily use hits the weekly cap first. So "great on Monday, throttled by Wednesday" is a pacing outcome, not an account problem.
And one counter-intuitive saver: letting a long session continue is usually cheaper than closing it and starting over, because a restart pays the context-reading cost again.
In practice, first-time users often exhaust a Pro plan within the first week and conclude the tool is unusable. The real cause is scope: pointing it at one directory with one verifiable goal costs an order of magnitude less than letting it wander a repository.
4. How the month gets paid for, and what you keep
Crypto only: USDT and USDC plus native TRX, SOL and TON, across the eight networks shown at checkout. No card, no PayPal, no bank transfer, so the "declined card" problem does not exist here. One invoice carries one network and one exact amount, and the network fee sits on top, paid by the sender.
No email is sent, ever. Your record is the order number and the transaction hash, and the hash is publicly verifiable. Activation needs an identifier, not a login: the Claude Organization ID, the org_ value from Settings, then Account. Never a password, never a 2FA or recovery code.
On region: nothing here is written as a workaround. If the vendor asks for identity verification, you go through the vendor's own flow — the verification guide says what that involves.
The order that works
- Complete the browser login and confirm the account is healthy, not restricted or mid-appeal
- Run a ten-minute continuous task to confirm the link holds
- Start with a narrow, verifiable task: "make this failing test pass"
- Watch two or three days of real consumption and see which ceiling you actually touch
- Then decide on a heavier tier, or spread the heavy work across days
How the limits work is covered in the rate-limit guide and the weekly cap explainer; why a long session costs less than a restart is here; which tier fits is in the 5x vs 20x comparison.
Common questions
Do I need a foreign card to pay for the plan?
No. Payment is crypto only — USDT, USDC, TRX, SOL and TON across eight networks. One invoice, one network, one exact amount, fee on top.
What do you need from my account?
Only the Organization ID copied from Settings, then Account, prefixed org_. Never a password, 2FA code or recovery code, and there is nothing to redeem on the vendor site.
The CLI says 401 after a successful browser login. Why?
Usually two different browser sessions, or the credential written under another user directory. Sign out, close all terminals, keep one browser on the plan account, then log in again.
I hit the limit after two days. Is the plan too small?
Not necessarily. Bursts drain the roughly five-hour window; steady use hits the weekly ceiling. Watch three days of real use before buying a heavier tier.
Do I get an email receipt?
No mail is sent at all. The order number and the on-chain transaction hash are the record, and refunds return to the address you sent from.
Guides
All guides →Claude Max 5x vs 20x: which tier you actually need
The difference is not quality — it is how many times you can hit the wall before lunch. A practical way to size your plan.
How to pay for Claude from mainland China (2026)
Four routes, ranked by how often they actually work — and the two that quietly cost you the account.