I suggest that stellar coalescence in mid-size protoclusters ($M {\sim} 10^{3.5}$–$10^{4.5} M_{\odot}$) is a possible scenario for the formation of ultraluminous X-ray sources (ULXs). More massive super-star-clusters are not needed, since the most likely ULX mass range is only ${\sim} 30$–$200 M_{\odot}$; in fact, they are very rarely found at or very near ULX positions. Protostellar envelopes and gas accretion favour captures and mergers in dense cores of embedded clusters. Moreover, protoclusters with masses ${\sim} 10^{3.5}$–$10^{4.5} M_{\odot}$ are likely to disperse quickly into loose OB associations, where most ULXs are found. Sufficiently high protostellar density may be achieved when clustered star formation is triggered by galaxy collisions and mergers. Low metallicity may then be necessary to ensure that a large fraction of the stellar mass ends up in a black hole. In this scenario, most ULXs are naturally explained as the extreme end of the high-mass X-ray binary population.