In Web2.0, OpenAPIs (such as Google Map, Flickr and Amazon S3) are considered to be among the most important and vital building blocks for the mashups that combine data and services provided by third parties through OpenAPIs, as well as internal data sources owned by users to aggregate user values and promote user innovation. But as a growing number of OpenAPIs are available on the Internet, the developer and user faces the challenge of dynamic complexity in finding and integrating the right set of OpenAPIs. In this paper, we propose a framework to address the problem of self-adaption in the Internet-scale mashups. First, we design a repository for storing OpenAPIs with Multi-granularity and then introduce QoS metrics measuring the potential impact of OpenAPI properties. We further encapsulate different OpenAPIs whose functionality is synonymous, but with different providers, communication protocols, data formats and parameters. Ultimately, we provide a ranking-style algorithm exploiting the encapsulated interface and QoS characteristics to dynamically select the right OpenAPI in runtime. In short, we propose a QoS-based mashup platform that can facilitate the construction of mashup and improve users' experience. The process is transparent and automatic to the users'. Experimental evaluation demonstrates the effectiveness of our approach.