Air　conditioners　in　residential　buildings　often　rely　on　local　feedback　control　to　regulate　the　indoor　temperature,　where　the　unit　is　cycled　on　and　off　by　comparing　the　actual　temperature　with　the　temperature　set-point.　Model　Predictive　Control　(MPC)　is　a　technique　that　makes　use　of　building　models,　disturbance　forecasts,　and　optimization　algorithms　to　enhance　building　performance.　MPC　has　demonstrated　its　potential　in　reducing　building　energy　consumption　and　operating　costs;　however,　it　has　seldom　been　applied　in　residential　buildings　because　of　the　high　implementation　costs,　complicated　hardware　configuration,　and　insufficient　data.　In　this　study,　we　developed　and　tested　a　MPC　controller　in　a　real　apartment　aimed　at　reducing　the　operating　costs　based　on　the　local　Time-of-Use　(ToU)　utility　structure.　We　developed　a　feasible　and　easy-to-use　hardware　configuration　as　well　as　a　cloud-based　software　deployment.　The　MPC　controllers　were　implemented　and　evaluated　through　one-month　field　testing　in　a　bedroom　and　living　room　of　an　apartment　in　a　residential　building.　The　field　test　showed　that　the　MPC　controller　saved　22.1%　for　the　bedroom　and　26.8%　cooling　costs　for　the　living　room,　compared　to　conventional　rulebased　control　(RBC).　We　further　confirmed　the　robustness　of　the　MPC　controller　as　the　field　test　showed　that　the　MPC　controller　is　able　to　outperform　the　conventional　RBC　controller,　even　if　the　model　is　not　accurate.　Our　work　provides　a　practical　MPC　solution　that　can　unlock　the　load　flexibility　potential　of　residential　buildings,　which　creates　a　win-win　solution　for　the　households　and　the　grid:　saving　utility　costs　for　households　and　helping　the　grid　to　manage　the　balance　between　supply　and　demand.(c)　2023　Elsevier　B.V.　All　rights　reserved.