Mobile agents have the potential to offer benefits, as they are able to either independently or cooperatively move throughout networks and collect/aggregate sensory data samples. They are programmed to autonomously move and visit sensory data stations through optimal paths, which are established according to the application requirements. However, mobile agent routing protocols still suffer heavy computation/communication overheads, lack of route planning accuracy and long-delay mobile agent migrations. For this, mobile agent route planning protocols aim to find the best-fitted paths for completing missions (e.g., data collection) with minimised delay, maximised performance and minimised transmitted traffic. This article proposes a mobile agent route planning protocol for sensory data collection called MINDS. The key goal of this MINDS is to reduce network traffic, maximise data robustness and minimise delay at the same time. This protocol utilises the Hamming distance technique to partition a sensor network into a number of data-centric clusters. In turn, a named data networking approach is used to form the cluster-heads as a data-centric, tree-based communication infrastructure. The mobile agents utilise a modified version of the Depth-First Search algorithm to move through the tree infrastructure according to a hop-count-aware fashion. As the simulation results show, MINDS reduces path length, reduces network traffic and increases data robustness as compared with two conventional benchmarks (ZMA and TBID) in dense and large wireless sensor networks.