Mobile autonomous agents, particularly robots, are becoming commonplace in many application domains, e.g., search and rescue, demining, agriculture and surveillance. These so-called Mobile Cyber-Physical Systems (M-CPS) allow relocating sensors and actuators dynamically in the environment to improve some global performance metric. However, the necessary agents cooperation is hindered by their heterogeneity, dynamic communication links and network topology, and by an error-prone communication channel. This chapter focuses on the networking and middleware support for a class of M-CPS and proposes the Reconfigurable and Adaptive TDMA protocol to handle data transmission under communication-related uncertainties, and the RTDB shared memory middleware to provide seamless data access across the cooperating agents. This combination of network protocol and middleware layer is particularly suited to support state sharing among agents in dynamic teams. Several experimental results confirm the advantage over other potential options. Moreover, this protocol and middleware combination have been thoroughly validated in demanding operational scenarios namely in robotic soccer teams from RoboCup Middle-Size League, which exhibit the typical requirements and constraints of M-CPS.