Why are you saying it doesn't provide predicted probabilities?
Like any other predictive model it gives the predictions.
Recall that a XGBoost model is a set of iterative Decision Trees (one Decision Tree built on the residual of the "previous" Decision Tree). In a single decision tree model, the predicted probabilities (assuming your target is binary) is the proportion of the event in the leaf where the analyzed case falls based on the tree path defined by its characteristics. It might be more elaborate, but conceptually XGBoost will give the prediction probability by following the paths of the different fitted Decision Trees and looking at the proportion of events in the final leaf reached.
What software are you using and what outcome are you obtaining with your XGBoost model?