Home → Guides → Pointing Claude Desktop and Codex at APICLAN
Desktop apps have no base-URL box in their own settings. CC Switch writes that configuration for you. Here is exactly what to enter, and the two things that account for almost every failed setup.
You need an account with balance on it and an API key. Create the key in the console under API Keys: pick the vendor family you are going to call — Anthropic for the Claude desktop app, OpenAI for Codex — and then the key group you want to spend from. The group sets your rate, and the pricing page lists what each model costs.
A key belongs to exactly one group. A key created for text models cannot call image models, and the other way round; that comes back as a 404 on the model rather than a permission error. If you need both, create two keys. The authoritative list of what a given key can call is always GET /v1/models sent with that key.
Every request has to arrive at a path starting with /v1. Whether you type /v1 into the client depends on the client, because some append it themselves and some do not. This one detail accounts for most of the “it will not connect” reports we receive.
| Client | Request address to enter |
|---|---|
| Claude Desktop (via CC Switch) | https://apiclan.us/v1 |
| Codex / ChatGPT desktop app | https://apiclan.us/v1 |
| OpenAI SDK, Cherry Studio, Chatbox, LobeChat, LM Studio | https://apiclan.us/v1 |
| Cursor, Cline, Roo Code | https://apiclan.us/v1 |
| Claude Code | https://apiclan.us |
| Anthropic SDK | https://apiclan.us |
Get it backwards and the request 404s at the edge, which means it never reaches the gateway and never appears in your usage log — so an empty log is itself the symptom. Never leave a trailing slash on the address.
Codex has been merged into the ChatGPT desktop application. On Windows, install ChatGPT from the Microsoft Store; on macOS, download it from OpenAI's site. The command-line Codex takes the same settings.
In CC Switch, add a provider for Codex, paste your key, and enter https://apiclan.us/v1 as the request address. Enable that provider, then quit Codex completely — not just close the window — and start it again. Configuration is read at launch, so a running instance keeps the old settings and it looks as though nothing happened.
If you would rather edit the configuration file by hand than use CC Switch, the base URL and where the key goes are covered in our quickstart.
Install the Claude desktop app from claude.com, then set the provider up in CC Switch:
https://apiclan.us/v1 as the request address — the client appends /messages to it. No trailing slash.If CC Switch offers you a choice between connecting directly and mapping models, read the next section before choosing. That setting is what decides whether the model picker works at all.
A banner saying the connection works but the provider rejected the request is almost always about the model name, not the key. The desktop app sends its own built-in model identifiers, and a gateway can only serve the models its upstream accounts actually provide.
The fix is model mapping rather than a direct connection: map the app's entries onto names this gateway serves. Connecting directly only works when the provider accepts the exact identifiers the app sends.
Send GET /v1/models with your key to get the list you can map onto. That list is built from your key's own group, so it is the real answer for your account rather than a general catalogue.
Nothing is overwritten permanently. In CC Switch, enable the official provider again and restart the client; the desktop app goes back to your subscription. You can keep both configured and switch whenever you like — balance here is prepaid and does not expire, so a setup you are not using costs nothing to keep.
/v1, and remove any trailing slash.Still stuck? Email support@apiclan.us with the model name that failed and roughly when you tried it. A person reads it.
No subscription, no monthly minimum, no sales call. Top up with USDT and spend what you use — 1 USDT gives you 2 credits of API balance.
Read the 30-second quickstartPrices quoted on this page are regenerated automatically from live billing data. Third-party terms are quoted from that party's own published documentation.