Estimating p-value for individual data points when performing anomaly detection?

I'm working on a time series anomaly detection project. Unfortunately, when I simply look for observations outside the confidence bounds Prophet estimates, there are many many false positives. I came across [this paper](https://openreview.net/pdf?id=NvN_B_ZEY5c) which proposes a way to control for false positives during online anomaly detection and it looks very promising but I need to get a p-value for each observation in the time series (the null hypothesis being that the latest observation is not anomalous).

Is there are clean way to obtain this through any of your public or private methods? Or what would you recommend?