Context. Protoclusters are progenitors of galaxy clusters and they serve as an important key in studies of how halo mass and stellar mass assemble in the early universe. Finding the signposts of such overdense regions, such as bright dusty star-forming galaxies (DSFG), is a popular method for identifying protocluster candidates. Aims. Hyperluminous infrared galaxies (HLIRGs) are ultramassive and show extreme levels of dusty star formation and black hole accretion that are expected to reside in overdense regions with massive halos. We study the Mpc-scale environment of the largest HLIRG sample to date (526 HLIRGs over 26 deg 2 ) and we investigate whether they are, in fact, predominantly located in overdense regions. Methods. We first explored the surface density of Herschel 250 µm sources around HLIRGs and made comparisons with the corresponding values around random positions. Then, we compared the spatial distribution of neighbors around HLIRGs with their counterparts around randomly selected galaxies using a deep IRAC-selected catalog with good-quality photometric redshifts. We also used a redshift-matched quasar sample and submillimeter galaxy (SMG) sample to validate our method, as previous clustering studies have measured the host halo masses of these populations. Finally, we adopted a friends of friends (FoF) algorithm to look for (proto)clusters hosting HLIRGs. Results. We find that HLIRGs tend to have more bright star-forming neighbors (with 250 µm flux density >10 mJy) within a 100 projected radius (∼ 0.8 Mpc at 2 < z < 4), as compared to a random galaxy at a 3.7σ significance. In our 3D analysis, we find relatively weak excess of IRAC-selected sources within 3 Mpc around HLIRGs compared with random galaxy neighbors, mainly influenced by photometric redshift uncertainty and survey depth. We find a more significant difference (at a 4.7σ significance) in the number of Low Frequency Array (LOFAR)-detected neighbors in the deepest ELAIS-N1 (EN1) field. Furthermore, HLIRGs at 3 < z < 4 show stronger excess compared to HLIRGs at 2 < z < 3 (0.13 ± 0.04 and 0.14 ± 0.01 neighbors around HLIRGs and random positions at 2 < z < 3, respectively, and 0.08 ± 0.04 and 0.05 ± 0.01 neighbors around HLIRGs and random positions at 3 < z < 4, respectively), which is consistent with cosmic downsizing. Finally, we present a list of 30 of the most promising protocluster candidates selected for future follow-up observations.