A hybrid numerical scheme based on a probabilistic method along with a classical domain decomposition is proposed for solving numerically linear elliptic boundary-value problems. Full decoupling can be accomplished by computing a few values of the solution inside the domain by Monte Carlo or quasi-Monte Carlo techniques, and interpolating at the nodal points where the solution has been obtained previously. Thus, this method appears to be fault-tolerant as well as suited for time decomposition. Some examples are shown to illustrate performance and scalability.