muav_gui_mcp

The server offers powerful, low-level control over robotics systems (UAVs, AGVs, ROS) via an AI agent. Tools like `send_command`, `ros_service_call`, and `pub_ros_topic` accept `z.any()` or `z.object({}).passthrough()` for their payload schemas. This lack of strict input validation for critical commands, particularly for ROS messages, means that if the AI agent or the MCP server is compromised, an attacker could send arbitrary, unvalidated commands to physical robots or the backend system, leading to unintended behavior, data corruption, or system disruption. The security of this MCP server is highly dependent on the security of the external Multi-UAV GCS API it connects to. Raw JSON responses from the backend are exposed, which could reveal sensitive data if not properly access-controlled by the GCS itself.