This paper studies the problem of online crowdsourced delivery for urban parcels using private cars under time-dependent travel times. We consider some complex and practical cases, such as time-dependent travel times, private car detours and distribution, and retail stores as delivery or transfer nodes. The online delivery problem is formulated as a mixed-integer programming model to maximize the total profit. For this, we design an improved adaptive large neighbourhood search to address the problem. The benchmark instances verify the effectiveness and applicability of the improved ALNS. Then, the instances of large-scale delivery network using the main urban area of Dalian as an actual scenario are experimented. Numerical experiments showed that well-planned online crowdsourced delivery can respond to service requests in a shorter time, and the profit is higher than FlashEx. Sensitivity analysis results show that the maximum detour coefficient is only increased within a certain range, which is helpful to improve the matching rate and total profit. Moreover, increasing the number of signed private cars can not only increase the direct delivery rate but also the matching rate and total profit.