Earthquake Early Warning Systems (EEWS) are often challenged when the earthquakes occur outside the seismic network or where the station density is sparse. In these situations, poor locations and large alert delays are more common because of the limited azimuthal coverage and the time required for the wavefield to reach the minimum number of seismic stations to issue an alert. Seismic arrays can be used to derive the directivity of the wavefield and obtain better location. However, they are uncommon because of the prohibitive cost of the sensors. Here, we propose the development of an array-based approach using mini-arrays of low-cost Microelectromechanical Systems (MEMS) accelerometers and show how they can be used to improve EEWS. In this paper, we demonstrate this approach using data from two MEMS Accelerometer Mini-Arrays (MAMA) deployed at University of California Berkeley and Humboldt State University. We use a new low-cost ( <U.S. $150) Data Acquisition Unit and solve for the back azimuth of seven events with magnitudes ranging from Mw 2.7 to 5.1 at distances of 5 km to 106 km.