Tectonic information available from remote seismometers is compromised by uncertainties in epicenter and depth, which often prevent earthquakes being associated with faults observable in morphologic data. Here we show that careful study of regional recordings can achieve reasonable position and, for the larger events, also depth estimates. We illustrate this with swarms totallying 90 earthquakes of 3.6 mb to 5.5 Mw recorded at coastal stations of the Brazilian Seismographic Network, GSN, GEOSCOPE and Cape Verdes. Seismicity occured in 4 swarms, firstly in 2012 with 7 events in the valley floor and eastern wall near the ridge-transform intersection (4.05º-4.2ºN). Next, five events occurred in 2014, with three in the inner floor (4.8º-4.9ºN) and two under the valley wall (4.7º-4.8ºN). A 70-earthquake swarm in 2016 occurred in the 4.4º-4.8ºN valley floor, in both walls and floor, involving slip on different faults. The last 8 events in 2019 were located outside the median valley near a volcanic seamount. We estimated focal depths of the strongest events with Mw > 5.4 using waveform modeling and focal mechanims reported by the Global Centroid-Moment-Tensor Project. Our best estimated hypocenters lie 5-8 km beneath the seafloor, in keeping with maximum depths typically found with OBSs on the MAR. In contrast, deeper events (10 km) have been found in an OBS experiment around an active detachment fault elsewhere. Multibeam sonar data from the area do reveal a detachment fault surface (“megamullion”), but it ends 10-15 km from the median valley floor, suggesting that it is probably inactive. Although our uncertainties still do not allow the event depths to be discriminated from 10 km, our best estimates are more compatible with shallower normal faulting. Over time, repeating this exercise with many such datasets and comparing with morphologic data should help to resolve the incidence of deeper ruptures associated with detachment faults.