The combined use of insulin and glucagon in the diabetes therapy has been proposed as a robust alternative to the classical insulin only approach relieving the patient of the psychological burden related to the risk of falling into hypoglycemia. The standard setup usually considered is based on two pumps controlled manually or, in future, automatically. While the results so far are very promising, the device complexity represents a serious disadvantage. Against this background, this paper examines the possibility to use a pump for insulin, but only a pen for glucagon, which becomes so in some way a safety net, used only occasionally. This paper uses an existing simulation model for the human glucoregulatory system to develop a model predictive controller for this setup and analyzed the effect of errors in the carbohydrate (CHO) estimations for the meal announcements, the most common cause of hypoglycemia. In-silico tests using the simulation model and data of two patients allow a proof of concept, showing that the risk of hypoglycemia can be removed and the glucagon pen injection becomes necessary only if the overestimation is sufficiently large. The amount of infused insulin and the time spent in hyperglycemia is hardly affected by these glucagon injections, but a higher or earlier insulin delivery is possible and increases the time in the euglycemic range. Thus the proposed method can retain most advantages of the dual pump approach while being more patient friendly. It is expected to encourage diabetics to target a better BG level without the usual safety margin as the glucagon availability takes over this safety function.