How to Use Claude Code Reliably in Mainland China: My Alternative Setup with cc-switch and MiniMax

Technology and Creation AI
1

After trying so many CLI tools—whether it’s GPT’s Codex or Google’s Gemini CLI—I’ve found myself coming back to Claude Code from Company A. It’s hands-down the CLI AI programming tool I feel most comfortable with, mainly thanks to Claude’s solid capabilities.

But for those of us based in mainland China, using Claude reliably has already become something of a luxury. To make things worse, on April 4th, Company A announced that Pro/Max plan quotas can no longer be used for third-party apps, explicitly calling out many OpenClaw setups.

So the problem is pretty straightforward. If you’re in China and want to use Anthropic’s official network stably, it’s far from easy. On top of that, third-party solutions have become increasingly unstable, with rising costs and no guarantee you’ll get all official features—or even a stable connection.

After going back and forth, I settled on a compromise. Instead of fighting with the official Claude API or relying entirely on third-party proxy services, I decided to keep using the Claude Code workflow, but switched the backend to a Chinese model compatible with the Anthropic protocol. In this post, I’ll explain how I did it: cc-switch + MiniMax.

image

Third-Party Comparison

I want to give you a straightforward comparison. If you can accept the pros and cons below, it’s fine to keep using third-party channels or a stable official setup:
Route Advantages Issues
Official Claude Network Full model capabilities, best experience Stability always an issue for users in mainland China
Third-party proxy/service Quick start, lots of community guides Frequent rule changes, extra charges becoming the norm
Compatible API replacement Controllable, you’re in charge long-term Model ability not equal to original Claude

Why MiniMax?

Really for just two reasons:

  1. MiniMax now officially provides an Anthropic API-compatible interface—well-documented and easy to integrate.

  2. Cheap pricing, with plenty of allowance.

I did consider GLM, but access is a scramble and people say it responds slowly, so I passed. I just bought the official Token Plan pack for 98 RMB/month—figured I’d give it a try anyway.

image

Claude Code Installation

This article won’t go into every detail of installing Claude Code, but here are the basic commands and how to check your install.

On Mac, Linux, or WSL:

curl -fsSL https://claude.ai/install.sh | bash
claude --version

Windows PowerShell:

irm https://claude.ai/install.ps1 | iex
claude --version

As long as your terminal recognizes claude, you’re good to go.

If you want full details on system requirements, paths, and platform-specific stuff, check out:

Further reading: Claude Code Installation Guide: Mac, Windows, Linux from Scratch

CC-switch Installation & Integration

The MiniMax official docs already explain linking with Claude Code pretty well, but here’s my own process for reference:

  1. Go to the MiniMax console and create an API Key.

  2. Clear out old environment variables: ANTHROPIC_AUTH_TOKEN and ANTHROPIC_BASE_URL

  3. Install cc-switch. I recommend just grabbing the right system version directly from its releases page.

If you’re on macOS, you can run:

brew tap farion1231/ccswitch
brew install --cask cc-switch
  1. Open cc-switch and create a new MiniMax provider, filling in the following key fields:

  • Provider: MiniMax

  • API Key: your MiniMax key

  • Base URL: https://api.minimaxi.com/anthropic (be sure to use the correct site—international or domestic—depending on where you purchased)

  • API Format: Anthropic Messages

  • Model: MiniMax-M2.7

image

  1. If you don’t sign in with an official account and directly change the provider, I recommend updating the onboarding status in ~/.claude.json:

    {
    “hasCompletedOnboarding”: true
    }

  2. Before you jump into real projects, test it in an empty directory first:

    claude

If everything’s set up, Claude Code should launch and you’ll see the model set to MiniMax instead of Claude.

image

Risk tip: The most common issue is not the Key itself, but leftover old settings or using the wrong base URL (since there are international and domestic sites). Sometimes, even if the UI says MiniMax is selected, requests still go to the old address. Try quitting and reopening the console or use logout to log out and sign in again.

Summary & Differences

This approach is what I’m using now, but I won’t hype it up as a “perfect Claude replacement.” Claude Code shines partly because the CLI is designed well, but ultimately the Claude model itself is great. Here, the shell is still Claude Code, but the underlying model is now MiniMax.

In short:

  • For everyday code Q&A, reading files, and simple logic tweaks, this workflow works just fine.

  • For long-term, stable work in China, it’s a lot more reliable than gambling on an official account. For some companies or individuals, using a domestic large model might be the only option.

  • But for complex repo understanding, long chain-of-thought reasoning, or agent tasks requiring fine detail, the experience definitely falls short of original Claude. That’s my honest assessment so far.

I still want to try integrating with GLM at some point, but since my MiniMax subscription hasn’t run out, I’ll stick with it for now, and buy more if I need to later. Mainly I’m just curious how far China’s domestic large models have come.

4 Likes
2

Looking back

3

Study and learn

1 Like
4

Learned it

5

Thanks for sharing! Learning :xhj001:

6

Check out my latest project for unlimited free access—it just launched this morning, so hurry and sign up!