Medium access control (MAC) layer scheduling is an intelligent solution to protract the lifetime of energy harvesting wireless sensor network (WSN) which is challenging issue in recent times. Self‐adaptive MAC protocol was working upon reinforcement learning. However, this decentralized adaptive scheduling introduces collisions and degrades the overall network performance. To fix this problem, this paper proposes novel semi‐synchronized MAC approach for energy harvesting WSN. The major objective of this work is to improve energy efficiency of the network by minimizing energy consumption which is the major issue in WSN. Proposed WSN network is modeled with dual mobile sinks with circular moving path and static sensor nodes to monitor the environment. Our novel MAC approach is initiated with unequal cluster formation by correlation based unequal clustering scheme in which multi‐objective elephant herding optimization (CUC‐MEHO) algorithm is incorporated for cluster head (CH) selection. In clustered network, all sensor nodes follow semi‐synchronized adaptive active period MAC scheduling (S2A2MAC) protocol. In this work, S2A2MAC operates upon hidden Markov model (HMM) which controls the active time period of a node in an adaptive manner. Energy efficient hop‐by‐hop intra‐cluster routing is empowered by Bayes rule enabled perfect matching (BR‐PM) algorithm. Before route selection, all candidate nodes are regarded into perfect match, partially match, and no match nodes. Furthermore, emergency data is identified by CH based on flag value and transmitted without any time delay. For emergency data transmission, extremity aware routing (EAR) algorithm is presented and EAR algorithm is functioning on hopcount and distance criteria. Finally, proposed network is evaluated by extensive simulations in NS‐3.26. Simulation upshots show better performance in percentage of dead nodes, network lifetime, energy consumption, PDR, and delay.