Fiber-reinforced cementitious matrix (FRCM) is a new-era material for strengthening concrete and masonry structures. Unlike Fiber reinforced polymer (FRP), it has excellent substrate bond properties. However, FRCM shows intra-fiber slip and Fiber-matrix debonding as its major failure modes, effectively reducing its ultimate load-carrying capacity. It has led to broad research scope, and researchers have used permutations and combinations of various types of fibers, matrices, coatings, and substrates to improve the tensile and bond properties of the FRCM composites. Additionally, other experimental parameters, such as rate of loading and experimental setup (clevis, hydraulic or mechanical), affect the tensile and bond properties of FRCM composites. Complex failure modes and parameters make tensile and bond capacity prediction very difficult. Previous attempts by researchers showed the effectiveness of machine learning in predicting the bond capacity of the FRCM composites. Artificial neural network (ANN) based models have also been used successfully to predict the flexural capacity of FRCM-strengthened masonry panels. This study develops a unified machine learning-based model to predict bond and tensile capacity using complex tree-based algorithms such as extra tree regressor, categorically boosted trees, extreme Gradient boosted trees, etc. The authors extracted 603 data points from the literature to develop and validate the model; this includes 345 data points from tensile testing and 258 data points from FRCM-concrete bond testing. It is found that the developed tree-based model predicts the ultimate tensile and bond capacity with about 93% and 95% overall accuracy, respectively.