When confronted with an unexpected turn of events or feeling spontaneous, travelers find themselves in need of on-the-spot planning assistance. We present Crowdcierge, a real-time crowd-powered trip planning application. Crowdcierge is capable of both planning new trips and re-planning on the fly based on unstructured natural language input. By using the retainer model, synchronous crowd collaboration, and crowdware, workers are quickly recruited to work together to accurately tag key ideas in a planning mission, plan the itinerary, and re-plan in response to problems that arise during the trip. This paper presents the design of each crowd task in Crowdcierge and preliminary results from the tagging task.