Multi-agent systems (MAS) applied to Embedded Systems enable cognitive agents to act in the physical world. However, the application of these systems has been little explored to automate communication during crisis events. With this approach, it would be possible to help collect real-time data and deploy rescue forces in risky locations. This paper describes a decentralized, proactive, and agent-based communication model. To validate the proposal, we applied our approach to a physical prototype of smart city devices. In this scenario, we include the interaction of different MAS in a simulation’s flood scenario. The results show the architecture’s effectiveness in collecting real-time data, providing a reliable, low-cost way to integrate citizens and governments during crises. The architecture is a promising approach for crisis management applications that takes advantage of the autonomous behavior of agents in MAS, improving the obtaining of subsidies to feed systems in smart cities and assisting decision-making.