Most active depth sensors sample their visual field using a fixed pattern, decided by accuracy, speed and cost trade-offs, rather than scene content. However, a number of recent works have demonstrated that adapting measurement patterns to scene content can offer significantly better trade-offs. We propose a hardware LIDAR design that allows flexible real-time measurements according to dynamically specified measurement patterns. Our flexible depth sensor design consists of a controllable scanning LIDAR that can foveate, or increase resolution in regions of interest, and that can fully leverage the power of adaptive depth sensing. We describe our optical setup and calibration, which enables fast sparse depth measurements using a scanning MEMS (micro-electro mechanical) mirror. We validate the efficacy of our prototype LIDAR design by testing on over 75 static and dynamic scenes spanning a range of environments. We also show CNN-based depth-map completion of sparse measurements obtained by our sensor. Our experiments show that our sensor can realize adaptive depth sensing systems.