A new algorithm has been developed in this study that can automatically determine regional moment tensor (MT) and its centroid depth with real‐time waveforms; it can do this within 2–4 min of an earthquake notice issued by the Central Weather Bureau (CWB). The program selects 3–7 BATS (Broadband Array in Taiwan for Seismology) stations based on three different strategies: best azimuthal distribution, highest signal‐to‐noise ratio, and shortest distances. The program then inverts MT solutions in parallel with various settings that include Moho depth of velocity models, frequency bands, and isotropic constraints. The optimal solution is determined via a search for the best waveform fit with an acceptable non‐double‐couple component that comes from the results of combinations of these inversion settings. Our new rapid MT report system greatly reduces the need for computational resources and avoids human judgments. By applying this full‐scanning approach on BATS (named AutoBATS), we redetermine the MTs for over 3000 regional earthquakes that took place between 1996 and 2016, the goal being to provide the most up‐to‐date possible MT catalog for the Taiwan area. Overall, the AutoBATS MTs are consistent with the Global Centroid Moment Tensors, with a mean difference in the Kagan angle of and of . Those focal mechanisms better illuminate the tectonic structures, which is a result of the significantly improved resolving ability for shallow () and deep () earthquakes. With the new regional MT catalog, we refine the relationship between moment and local magnitudes: for the Taiwan region.