Measurements of the cosmic redshift drift - the change in redshift of a source over time - will enable independent detection of cosmological expansion thanks to the immense precision soon reached by new facilities such as the Square Kilometer Array Observatory and the Extremely Large Telescope. We conduct the first ever redshift drift computation in fully relativistic cosmological simulations, with the simulations performed with the Einstein Toolkit. We compute the redshift drift over the full skies of 50 synthetic observers in the simulation. We compare all-sky averages for each observer - and across all observers - to the Einstein-de Sitter (EdS) model which represents the large-scale spatially-averaged spacetime of the simulation. We find that at $z\approx0.2$ the mean redshift drift across the sky for all observers deviates from the EdS prediction at the percent level, reducing to $\sim0.1\%$ by $z\approx 1$. However, fluctuations in the redshift drift across the sky are $\sim 10-30\%$ at $z\approx 0.1$ and a few percent at $z\approx 0.5$. Such fluctuations are large enough to potentially exceed the expected precision of upcoming redshift drift measurements. Additionally, we find that along 0.48% of the light rays the redshift drift becomes temporarily positive at very low redshift of $z\lesssim 0.02$. This occurs despite our simulation data being based on a matter-dominated model universe. By including a cosmological constant, we expect a slower growth of structures than in the leading-order EdS space-time, and this may reduce the anisotropy over the observers' skies, although we generally expect our results to hold as order-of-magnitude estimates. Redshift drift is arguably one of the most important measurements to be made by next-generation telescopes. Our results collectively serve as preparation for interpreting such a measurement in the presence of realistic cosmic structures.