When I did this I just used a custom token in visual studio and then looked at the logs in the provider's web UI.

I don't think they care very much about this. The encrypted blobs for reasoning models is a different matter.