This paper presents a new Multi Agent Collaborative Search Algorithm with Adaptive Weights (named MACSAW). MACS is a memetic scheme for multi-objective optimization which contains two kind of actions, the local actions and social actions. The former explore the neighborhood of some virtual agents and the latter push the individual towards the Pareto front. On the base of the latest version of MACS, MACS2.1, we improve the old algorithm from three direction. First, a new kind of utility function is introduced to enhance the convergence. Next, a new social action process which contains more operators and adaptive parameters is embedded in MACSAW. Finally, MACS2.1 lacks the weight vectors adjustment process which leads to diversity losing in some real problems and MACSAW adds it. Further, MACSAW is compared with some state-of-art algorithms and MACS2.1 on some standard benchmarks. It gets competitive results. Two real optimization problems is tackled and the results are analyzed in details.