China Claude Proxy Sharing Model | Generated by AI

Home 2026.03

Question: What is a “Claude Code mirror” service, especially in China? Specifically, the distribution model where it acts like a proxy/relay server — where one person buys a Max plan and shares it with many others?

Answer:

What Is a “Claude Code Mirror” in China?

Background: Why It Exists

Claude (and Claude Code) is officially blocked in mainland China. Anthropic does not support Chinese payment methods, Chinese IP addresses are often rejected, and starting from September 5, 2025, Anthropic officially announced a policy banning all “Chinese-controlled companies” from using Claude services, including API and Web products. This created a large grey market of workaround services.


What Is a “Claude Code Mirror”?

A Claude Code mirror (also called “CC mirror”, 镜像服务, or relay service) is a third-party proxy/relay platform that sits between Chinese users and Anthropic’s official servers. These are third-party platforms providing Claude Code mirror access for developers in China, offering an API-compatible service layer for stable access — with no official authorization from Anthropic.

The key mechanism: the mirror service intercepts your API requests and forwards them to Anthropic on your behalf, using servers hosted in compliant regions (outside China). Users simply change their ANTHROPIC_BASE_URL environment variable to point to the mirror instead of Anthropic’s servers directly.


The “Bus” / Shared Pool Model (拼车 / 合租)

Your “bus” analogy is spot-on. This is colloquially called 拼车 (carpooling) or 合租 (shared rental) in Chinese developer communities. Here’s how the business model works:

The operator’s side:

The user’s side:

The relay service infrastructure: Open-source tools like claude-relay-service act as a middleware layer that forwards requests while managing authentication, routing, and cost-sharing scenarios. The system is particularly useful for teams or communities that want to pool access or simplify integration with different AI backends.


Types of Services in This Ecosystem

There are several distinct models:

Type How It Works Risk Level
Official API reseller Buys Anthropic API credits, resells at markup Medium
Subscription sharing (Max pooling) One Max plan shared among many users High — violates ToS
Third-party reverse-engineered Uses technical exploits/unofficial endpoints Very High
VPN + direct access User configures VPN proxy themselves Lower, but still restricted

This is the critical part:

  1. Violates Anthropic’s Terms of Service — Sharing a subscription account or reselling API access without authorization is explicitly against the ToS. In the second half of 2025, multiple Claude reseller services shut down due to Anthropic’s strengthened risk control, causing significant losses for prepaid users.

  2. Account bans — Anthropic has multi-layer detection mechanisms checking IP addresses, DNS leak locations, browser fingerprints, and device fingerprints. Mainland China IP ranges are rejected, and data center IPs or shared VPN IPs are flagged as high risk.

  3. Operator disappearance risk — Over 60% of third-party Claude reverse engineering services that emerged in 2025 shut down or disappeared within 3 months of operation. Users who prepaid lose their money.

  4. Security risk — When you route all your code and prompts through a third-party relay, that operator can potentially log your code, API calls, and data.

  5. China policy — Third-party reverse-engineered services use technical loopholes to provide unofficial services, violating Anthropic’s terms of service, and can be banned at any time.


Legitimate Alternatives for Chinese Developers


Summary

A “Claude Code mirror” in China is essentially a paid relay proxy service that solves access and payment barriers. The “bus/pooling” model specifically refers to operators pooling many users onto one or a few accounts/API keys, billing users in RMB at a markup. While popular and practically useful, these services carry real risks: ToS violations, account bans, operator fraud, data security issues, and instability.

References:


Back Donate