Claude Desktop

The Claude desktop app uses the Anthropic Messages protocol, which Amux implements natively.

Recommended: let CC Switch do it#

The desktop app keeps its configuration in its own file, in a location that varies by system and is easy to get wrong by hand — CC Switch has a UI for it. Select Claude Desktop and add a provider:

Quit the desktop app completely and reopen it after switching — it reads configuration at startup, and minimising to the tray is not quitting.

By hand#

Where the file lives#

What to change#

Point the base URL at https://gateway.amux.ai and swap in your Amux key.

Do not append /v1 — Anthropic clients build /v1/messages themselves, and the extra segment turns into /v1/v1/messages, which comes back 404.

Quit completely and reopen afterwards.

Choosing models#

Model names are canonical IDs (anthropic/claude-opus-5). Non-Claude models work too — openai/gpt-5.6, for instance — because protocol and model are two separate things, as long as that model's upstream serves the Anthropic protocol. Each model's page says so.

To pin a provider or choose a routing strategy, append a colon — see the API reference.

Troubleshooting#

Nothing changed#

The desktop app did not actually quit. On macOS closing the window is not quitting — use Cmd + Q. On Windows, check whether it is still in the tray.

404s#

The address has an extra /v1. Drop that segment.

Authentication failure#

Was the key copied in full (leading or trailing whitespace is the usual culprit)? Is it still enabled in the console? Does the workspace have balance?

Model unavailable#

Model IDs need the vendor prefix, and the model's upstream has to serve the Anthropic protocol — each model's page lists the protocols actually available for it.