This paper focuses on a Multi-Agent Federated Q-Learning framework for enhanced intrusion detection in Wireless sensor networks. The proposed approach incorporates Federated Learning to enable decentralized, privacy-preserving collaboration among sensor nodes. Each sensor node locally builds and updates a Q-table based on observed state-action pairs and sends these updates to a gateway node. The gateway aggregates the received Q-tables, refines the global Q-table, and disseminates it back to all sensor nodes, ensuring synchronized and adaptive defense mechanisms. This framework is implemented using a gateway node, three sensor nodes, and simulating a realistic WSN environment and presents a lightweight, scalable, and secure solution for intrusion detection in WSNs. Our system uses the WSN-DS dataset for evaluation, encompassing attack scenarios such as blackhole, grayhole, flooding, and TDMA and extended the dataset by incorporating replay and Sybil attacks. These new attack scenarios are synthesized using Generative Adversarial Networks to create a more comprehensive dataset. The model is then trained and tested to identify all six attack types, improving its robustness against a various cyber threat. Experimental results demonstrate the federated Q-Learning method achieves high detection accuracy while maintaining low overhead, making it usable for resource-constrained WSNs. This work extends an efficient and scalable method for collaborative intrusion detection, emphasizing both robustness and energy efficiency in distributed sensor networks.