The search for a physiologically appropriate control interface for prosthetic hands is an ongoing challenge in rehabilitation robotics. We proposed a myokinetic interface based on localizing magnets implanted in the residual muscles to monitor contractions and decode user intention. In the first human demonstration, we used a transcutaneous magnet localizer with a computational unit implementing magnet localization by acquiring synchronized magnetic field recordings from multiple acquisition units. These had to be arranged in fixed relative position and orientation, a constraint that substantially affected the system footprint. To overcome this limitation, here we present a modular system in which localization is implemented directly on each acquisition unit processor. Compared to the previous solution, this system achieves similar localization accuracy and precision with one or two magnets, while we devise future improvements in computation efficiency. This system significantly reduces the interface footprint and can be smoothly adapted to different anatomical districts.