MPI is a process to process collaboration via a message passing mechanism. Although this can be on a shared memory system it need not be. TBB and OpenMP (or Cluster OpenMP or other thread mechanism) is generally designed around a shared memory module. The two can be combined as a hybrid.
Usually you would code the OpenMP/TBB application such that outside the parallel sections the master thread runs the MPI message loop which then calls into routines that spawn multiple threads, performs the work, then upon completion, exits the parallel section and returns to the master threadMPI message loop. The MPI does not have a concept of having a thread within a spawned application.