Abstract: In order to realize rapid response to manufacturing needs and provide with the most ideal manufacturing cloud services(MCS),a dynamic clustering optimization approach for multi-granularity MCS is proposed.An analytical framework is given to organize manufacturing cloud services in the order of granularity scales through the decomposition of manufacturing needs,thus the cloud resource pool is filtered for the first time.A multi-objective model based on total cost,service quality and delivery time is constructed and solved by Epsilon-SPEA2 so that Pareto optimal solutions can be obtained.An improved TOPSIS technique is employed to assess the solutions which represent the clustering schemes,thus the best clustering scheme can be selected.A practical example related to a project in mould industry is given to validate its effectiveness and feasibility.