A system and method for identifying errors in a video conference conducted on a network are disclosed. A central device includes a central agent that detects an error during a video conference conducted on a network between two or more endpoints. The central agent distributes a capture agent via the network to a remote device associated with a collision domain that contains one of the two or more endpoints. The capture agent collects a plurality of media packets associated with the video conference in a capture file located in a storage medium associated with the remote device until a timer expires. When the timer expires, the capture agent communicates the capture file to the central agent at the central device.