In this paper, a cooperative search method, based on a multi-agent structure is developed to deal with the $k$-coloring problem. Three agents coordinate using an adaptive memory, a search agent, an intensification agent and a diversification agent. We use the results of a preliminary fitness landscape study to adjust the navigation strategy in the solution space and to fix the search parameters. Our method provides competitive results and it is fast when compared with best existing techniques on instances extracted from the second DIMACS challenge.