Online restaurant aggregators with integrated meal delivery networks have become more common and more popular in the past few years. Meal delivery is arguably the ultimate challenge in last-mile logistics: a typical order is expected to be delivered within an hour (much less if possible) and within minutes of the food becoming ready. We introduce a novel formulation for a meal delivery routing problem (in which we assume perfect information about order arrivals) and develop a simultaneous column- and row-generation method for its solution. The analysis of the results of an extensive computational study, using instances derived from real-life data, demonstrates the efficacy of the solution approach, and provides valuable insights into, among others, the (potential) benefits of order bundling, courier-shift scheduling, and demand management.