In this paper we introduce a semantic search technique based on ontological concept learning. We also present a prototype of a multi-agent system (MAS) that can handle semantic search and at the same time hide the search complexity from the user. MAS can handle distribution and decentralization of information at the expense of ontology diversity. In order to overcome the difficulty of communication between agents with diverse ontologies, we suggest integrating semantic search with concept learning to enable agents to learn concepts from each other and therefore understand each other better. Ontological concept learning helps an agent to understand new concepts from peer agents. We use social networks to manage communication between agents and to improve the learning process by resolving the conflicts that may occur during learning new concepts from several agents.