We calculated the desorption energy of MgH2 clusters using the highly accurate quantum Monte Carlo (QMC) approach, which can provide desorption energies with chemical accuracy (within similar to 1 kcal/mol) and therefore provides a valuable benchmark for such hydrogen-storage simulations. Compared with these QMC results, the most widely used density functional theory (DFT) computations (including a wide range of exchange-correlation functionals) cannot reach a consistent and suitable Level of accuracy across the thermodynamically tunable range for MgH2 clusters. Furthermore, our QMC calculations show that the DFT error depends substantially on cluster size. These results suggest that in simulating metal-hydride systems it is very important to apply accurate methods that go beyond traditional mean-field approaches as a benchmark of their performance for a given material, and QMC is an appealing method to provide such a benchmark due to its high level of accuracy and favorable scaling (N-3) with the number of electrons.