With the increasing development of Location-based services (LBS), the spatial data become accessible on the web. Often, such services provide a public interface which allows users to find k nearest points to an arbitrary query point. These services may be abstractly modeled as a hidden database behind a kNN query interface, we refer it as a kNN hidden database. The kNN interface is the only way we can access such hidden databases and can be quite restrictive. A key restriction enforced by such a kNN interface is the top-k output constraint - i.e., given an arbitrary query, the system only returns the k nearest points to the query point (where k is typically a small number such as 10 or 50), hence, such restriction prevents many third-party services from being developed over the hidden databases. In this paper, we investigate a interesting problem of "breaking" the kNN restriction of such web databases to find more than k nearest point. To our best knowledge, this is the first work to study the problem over the kNN hidden database. We investigate and design a set of algorithms which can efficiently address this problem. Beyond that, we also perform a set of experiments over synthetic datasets and real-world datasets which illustrate the effectiveness of our algorithms