Distributed energy resources like solar panels and wind turbines are increasingly replacing fossil fuel energy sources. Due to their natural fluctuations they cause imbalance in the power grid when their production does not overlap with demand. Thus, there is a need to re-think the power grid into a system where the demand could be constantly adjusted to follow the fluctuation of the electricity generation. Demand Response (DR) is a mechanism to shift the electricity consumption of end-use customers from times with a high wholesale market price or when the security of the system is threatened to other time periods through various economic incentives. In Europe more than 20% of the total energy consumption is used for heating, cooling, and domestic hot water of buildings. Heat pumps are predicted to become one of the main heating and cooling technology in the future. Since heat pumps can be powered by electricity, they are regarded as a great source of flexibility, especially when combined with building inertia and heat storage to be used in DR. The goal of this thesis is to assess the benefits and challenges of implementing a flexible control of heat pumps. Although this question has been addressed in the literature, most studies use simulations. Here, we take an applied approach by implementing a flexible controller on a pilot site of inhabited residential buildings in Switzerland. We developed an interface using Model Predictive Control (MPC), that was tailored to the particularities of each of the pilot site's buildings. After validating the system using simulation, we have investigated the implementation of the necessary infrastructure to realise DR on a pilot site. The implementation highlights technical difficulties related to the integration on an existing and heterogeneous system, with limited monitoring capabilities. The pilot site allowed us to demonstrate that we can supply DR services to a third-party electricity aggregator. Although the heat pumps were difficult to control because they have non-linear behaviour and long reaction times, we were able to react to DR calls from the aggregator. The work provides a comprehensive study of the effort required to implement MPC in existing residential buildings, and confirms that DR in combination with heat pumps is a promising set-up to provide flexibility in the future.