The fuzzy fingerprint vault has been proposed for protecting the fingerprint template by adding a number of redundant features. Recently, a correlation attack that finds the real features using multiple fingerprint vaults has been introduced by several researches and should be solved for practical implementations. To avoid this correlation attack, we propose the Match-on-Card system based on the fuzzy fingerprint vault where the fingerprint matching is performed by an in-card processor, not an external card reader. Since the vault information is not released from the smart card, it is difficult for an attacker to take multiple fingerprint vaults. However, because of the limited resources (i.e., processing power and memory space) of the smart card, integrating fuzzy fingerprint vault into it is still an open challenge. To overcome this limitation, we propose the configurable fuzzy fingerprint vault based on a geometric hashing. Experimental results show that our approach can obtain much less memory requirements (i.e., 50KB) and faster execution time (i.e., 1.5sec) than the previous ones (i.e., 277KB, 6.6sec) without a significant degradation of the verification accuracy.