The　task　priority　planning　problem　is　addressed　in　the　task　supervisor　of　null-space　behavioral　(NSB)　control　for　multi-agent　systems.　Traditional　methods　rely　on　pre-defined　logic-based　or　fuzzy　rules　to　adjust　task　priority.　In　this　work,　a　novel　task　supervisor　is　proposed　using　model　predictive　control　(MPC)　techniques.　At　each　sampling　instant,　the　task　priority　planning　problem　is　formulated　as　a　switching　mode　optimal　control　problem　(OCP),　which　can　be　solved　by　efficient　mixed-integer　optimal　control　algorithms.　The　optimal　task　priority　order　is　obtained　based　on　current　and　predictive　information　of　agents,　without　the　need　for　a　pre-defined　rule.　By　explicitly　introducing　slack　variables　into　constraints,　the　proposed　MPC　method　is　flexible　to　cope　with　dynamic　obstacles　in　unknown　environments.　Simulations　with　static　and　dynamic　obstacles　show　that　the　proposed　method　can　provide　significantly　better　control　performance　than　the　traditional　logic-based　method　using　less　priority　switchings.