Many algorithms compute shortest-path queries in mere microseconds on continental-scale networks. Most solutions are, however, tailored to either road or public transit networks in isolation. To fully exploit the transportation infrastructure, multimodal algorithms are sought to compute shortest-paths combining var- ious modes of transportation. Nonetheless, current solutions still lack performance to efficiently handle interactive queries under realistic network conditions where traffic jams, public transit cancelations, or delays often occur. We present MUSE, a new multimodal algorithm based on graph separators to compute shortest travel time paths. It partitions the network into independent, smaller regions, enabling fast and scalable preprocessing. The partition is common to all modes and independent of traffic conditions so that the pre- processing is only executed once. MUSE relies on a state automaton that describes the sequence of modes to constrain the shortest path during the preprocessing and the online phase. The support of new sequences of mobility modes only requires the preprocessing of the cliques, independently for each partition. We also aug- ment our algorithm with heuristics during the query phase to achieve further speedups with minimal effect on correctness. We provide experimental results on France’s multimodal network containing the pedestrian, road, bicycle, and public transit networks.