Can’t Use Claude Code or Codex in China? Set Them Up with Domestic Models in 2 Minutes

Many developers in China struggle to use popular AI‑coding tools Claude Code and Codex because foreign accounts are unavailable or the official usage quotas are prohibitively expensive, and there is a risk of account bans.

Both tools support model switching, so the solution is to route their requests through a domestic LLM. The open‑source, cross‑platform desktop application CC Switch provides a visual interface to manage configurations for multiple AI‑coding tools and includes preset suppliers for models like DeepSeek, Qwen, Kimi, and others.

Install CC Switch

Download the appropriate installer from the CC Switch website or GitHub Releases. On macOS you can install with a single Homebrew command: brew install --cask cc-switch Windows users download the .msi package, and Linux users choose the .deb, .rpm or .AppImage variant.

Prepare a DeepSeek API Key

Register on the DeepSeek Open Platform, create a new API key on the API Keys page, and copy it immediately because it is shown only once.

Connect DeepSeek to Claude Code

Install Claude Code ( npm install -g @anthropic-ai/claude-code) and run claude to verify the login step. Then open CC Switch, select the Claude application, click “Add Supplier”, choose the preset DeepSeek , paste the API key, and keep the other fields at their defaults. The preset includes two model versions: DeepSeek‑V4‑Pro (stronger agent capabilities) and DeepSeek‑V4‑Flash . Optionally enable the 1M token mode for long contexts. After saving, enable the model and restart Claude Code; asking “What model are you?” should return a DeepSeek‑based response.

Connect DeepSeek to Codex

Install Codex CLI ( npm install -g @openai/codex) and run codex. Directly editing ~/.codex/config.toml to replace the OpenAI base_url with DeepSeek’s endpoint fails because Codex uses the OpenAI Responses API while DeepSeek implements the Chat Completions API . The mismatch causes 404 errors.

CC Switch solves this by providing a local routing proxy that translates Codex requests into the format DeepSeek understands. Enable “Local Routing” for Codex in CC Switch, add the DeepSeek supplier (same steps as for Claude), and turn on the routing switch. The proxy transparently forwards requests: Codex → CC Switch → DeepSeek → CC Switch → Codex. After saving, restart Codex; the model now reports as DeepSeek while still presenting the original Codex interface.

Verification and Reversal

For both tools, verify the switch by asking the model to identify itself. To revert, simply disable the local routing and restore the default configuration.

Using an open‑source tool and domestic LLMs reduces the cost to a fraction of the official pricing and eliminates the need for foreign accounts, making AI programming accessible in China.