Connecting AI Tools to Claude / GPT From China: Cursor, Cherry Studio, Cline, LobeChat, one-api Setup
Connect an OpenAI-compatible, CNY-fundable relay in Cursor, Cherry Studio, Cline, LobeChat and one-api: change base_url and key to call Claude / GPT / Gemini.
1. The shared principle: all these tools accept a custom OpenAI endpoint
Cursor, Cherry Studio, Cline, LobeChat and one-api all accept a 'custom OpenAI-compatible endpoint', so the integration logic is identical and needs three values: ① API address (base_url) https://cocodot.co/api/ai/v1; ② the API key created in your console; ③ a model code — mco-6 (Claude Opus), mog-6 (GPT), mgg-7 (Gemini 3 Flash) or deepseek-v4-flash. Below, where each tool takes those three values. One key, one balance, Alipay top-up, every tool covered, models switchable at any time.
2. Cherry Studio
In Cherry Studio's Model Providers settings, add a provider of the OpenAI type: API address https://cocodot.co/api/ai/v1, your key, then add models manually with IDs mco-6 / mog-6 / mgg-7 / deepseek-v4-flash. Save and pick a model under that provider in a conversation. Cherry Studio is a pure chat client with the smoothest basic chat experience — good for daily Claude / GPT use.
3. Cursor
In Cursor Settings → Models, enable 'Override OpenAI Base URL', set it to https://cocodot.co/api/ai/v1, enter your key, and add a custom model name (e.g. mco-6). Note: some of Cursor's advanced Agent / Composer features place demands on model capability — prove basic chat and completion first, and test complex agent modes on your real usage with a small top-up before scaling.
4. Cline / Roo Code (VS Code extensions)
In Cline or Roo Code settings, set API Provider to 'OpenAI Compatible', Base URL to https://cocodot.co/api/ai/v1, API Key to your key, Model ID to mco-6 or another code. Reminder: their agent modes (editing files, running commands) depend on the model's tool-calling ability, and support varies by endpoint — top up small, run your real task once, then decide whether to make it your main setup.
5. LobeChat
In LobeChat (self-hosted or client), under Language Models find OpenAI or 'custom AI service', set the API proxy address to https://cocodot.co/api/ai/v1, API Key to your key, and add mco-6 / mog-6 / mgg-7 / deepseek-v4-flash to the custom model list. It is an open-source chat UI, good for running your own multi-model chat site.
6. one-api / new-api (aggregation)
If you self-host one-api / new-api, add cocodot as an upstream channel: new channel, type OpenAI, proxy / Base URL https://cocodot.co/api/ai/v1, your key, and the supported model codes. Your aggregation layer gains a CNY-fundable OpenAI-compatible route for failover or traffic splitting.
7. Want Claude? Every tool above can call it
One cocodot key calls Claude Opus (mco-6), GPT (mog-6), Gemini (mgg-7) and DeepSeek, with Alipay top-up, metered billing, prices capped at list (see cocodot.co/pricing) and free switching for comparisons — to code with Claude, configure Cursor / Cherry Studio / Cline as above. As for the Claude Code CLI: it speaks Anthropic's own /v1/messages format, separate from the OpenAI-compatible endpoint — but cocodot provides an Anthropic-compatible endpoint for it, so two environment variables connect Claude Code to the real Claude (see our Claude Code guide). IDE or Claude Code, Claude runs on cocodot either way.
8. General pitfalls
① Model codes follow the console / docs and differ between vendors — do not copy from elsewhere; ② always top up small and complete one real call first, confirming latency and model before scaling — the most important habit in this field; ③ for complex agent / tool-calling usage, test your real task before relying on it; ④ base_url ends in /v1 — no less, no extra path.