In large-scale disasters, secondary disasters such as fires, landslides, and building collapses pose significant threats to evacuees. Therefore, it is crucial to accurately identify dangerous locations and construct appropriate evacuation routes that detour such locations. Even though the Internet and related communication service is an effective means to address this issue, existing communication infrastructure may become unavailable due to damage caused by the first strike. To ensure efficient evacuation guidance even in such situations, this study proposes an evacuation support system based on multi-agents utilizing unmanned aerial vehicles (UAVs) that do not rely on conventional communication infrastructure. The proposed system is composed of multiple agents that handle multiple devices with distinct roles, including search UAVs that assess the disaster area, guide UAVs that lead evacuees, relay UAVs that facilitate communication, and a server that constructs optimal evacuation routes. The agent system calculates the evacuation priority of each location based on the likelihood of secondary disasters, the number of evacuees present, and the distance to the safe places. In order to demonstrate the feasibility of the system, we conducted a simulation using NS-3. The results demonstrate that increasing the number of guide UAVs significantly improves the effectiveness of evacuation guidance.