The LoRa networks allow building low-cost longrange networks. The selection of LoRa access point locations in large scale networks requires taking into consideration the spatial distribution of clients, radio signal propagation and the limitation of the number of devices served by a single point of access. We propose a heuristic algorithm for the selection of access point locations within for a given set of candidate locations, based on gradient optimization. The proposed method allows selecting a sub-optimal set of locations which provide full coverage and take into consideration the capacity dimensioning based on spreading factor and expected channel utilization. The algorithm is evaluated in random topologies and in reallife scenario based on city-wide smart meters locations. The analysis shows that it outperforms a manual selection of access points locations, allowing to decrease the number of access points providing the full coverage.