We propose to use high-redshift long $\gamma$-ray bursts (GRBs) as cosmological tools to constrain the amount of primordial non-Gaussianity in the density field. By using numerical, N-body, hydrodynamic, chemistry simulations of different cosmological volumes with various Gaussian and non-Gaussian models, we self-consistently relate the cosmic star formation rate density to the corresponding GRB rate. Assuming that GRBs are fair tracers of cosmic star formation, we find that positive local non-Gaussianities, described in terms of the non-linear parameter, \fnl, might boost significantly the GRB rate at high redshift, $z \gg 6$. Deviations with respect to the Gaussian case account for a few orders of magnitude if \fnl$\sim 1000$, one order of magnitude for \fnl$\sim 100$, and a factor of $\sim 2$ for \fnl$\sim 50$. These differences are found only at large redshift, while at later times the rates tend to converge. Furthermore, a comparison between our predictions and the observed GRB data at $z > 6$ allows to exclude large negative \fnl, consistently with previous works. Future detections of any long GRB at extremely high redshift ($z\sim 15-20$) could favor non-Gaussian scenarios with positive \fnl. More stringent constraints require much larger high-$z$ GRB complete samples, currently not available in literature. By distinguishing the contributions to the GRB rate from the metal-poor population III regime, and the metal-enriched population II-I regime, we conclude that the latter is a more solid tracer of the underlying matter distribution, while the former is strongly dominated by feedback mechanisms from the first, massive, short-lived stars, rather than by possible non-Gaussian fluctuations. This holds quite independently of the assumed population III initial mass function.