Magnetic Resonance Imaging (MRI) scanners use strong magnetic fields to safely image soft tissues deep inside the body. They offer a unique tool for guiding therapies: images while patient is inside the scanner can localize diseased tissue and guide an intervention with high accuracy. This research controls MRI magnetic fields to wirelessly push millimeter-scale robots through vessels in the body, assemble them into tools, and provide targeted drug delivery or pierce tissue. This will directly impact healthcare, improving patient outcome by enabling unparalleled minimal invasiveness resulting in faster recovery, fewer side effects, and cost-effectiveness. This transformative toolset for multi-agent control will set the foundation for a wealth of medical therapies and surgical interventions. Using magnetic forces of clinical MRI scanners to steer miniature tetherless effectors through human bodies and combining with real-time imaging and operator immersion could transform the practice of minimally invasive interventions. This CPS will seamlessly integrate physical (scanner sensor/actuator, effectors, patient, operator) and cyber (world modeling, combined sensor and effector control, operator immersion). Work entails: (1) Portfolio of parametric effector designs that can be optimized to exploit the constraints of a given clinical procedure. (2) Toolbox of automatic controllers for MRI-based powering and steering of tetherless effectors in the body lumen, self-assembling them into tools, and precision therapy delivery or to pierce tissue. (3) Real-time MRI-based sensing of the physical world for imaging and tracking effectors and tissue. (4) Linked effector and MRI scanner control on-the-fly. (5) Visual/force-feedback human-robot interfacing. The work focuses on two effector classes: an MRI Gauss gun that stores magnetic potential energy released by a chain reaction when robots self-assemble, and an MRI pile-driver that converts kinetic energy from an enclosed sphere into impulses to tunnel into tissue. These approaches will be validated through analytical modeling, scaled hardware experiments, and experiments in clinical MRI scanners.