This paper presents a novel modularity-based consensus community detection (MCCD) algorithm that exploits the concept of consensus over N independent trials to generate robust communities and to aggregate marginal nodes to a single community. The algorithm is tested on LFR benchmark networks, a class of artificial networks with built-in community structure that can be made to reflect the properties of real-world networks. Preliminary results show that MCCD outperforms a single run of the original algorithm in terms of Normalised Mutual Information (NMI), number of communities and community size distribution, and provides an effective tool for community detection in real-world networks and a way to overcome the dependence on random seed of modularity-based algorithms.