I don't have any experience with MSIsensor. In our lab, we use MSIseq, which offers a decision tree model classifier. You can train a decision tree model or use the default one. It then classifies the tumor as "MSI-H" or "non-MSI-H".
Maybe this might help; from MSIseq's article: "Tumors in which ≥40% of the markers in a panel show somatic length mutations are generally termed MSI-high (MSI-H)19. Tumors in which no markers show length mutations are termed microsatellite stable (MSS). The remaining tumors are sometimes termed MSI-low (MSI-L). "
According to the original paper the MSI status should be called based on a MSIsensor > 3.5. In the literature different cutoffs have been used: 4 (here and here), or even up to 10 (here and here) depending on the sensitivity/specificity trade-off of interest and the biological context. If you don't have any experimental data on which you can test your cutoff, I assume these values could be a good starting point.