The proliferation of web sites that disseminate fake news is a growing problem in our society. Not surprisingly, the problem of identifying whether a web page contains fake news has attracted substantial attention. However, the problem of discovering new sources of fake news has been largely unexplored. Timely discovery of such sources is critical to combat misinformation and minimize its potential harm. In this paper, we present an automatic discovery system that proactively surfaces fake news domains before they are flagged by humans. Our system operates in two-steps: first, it uses Twitter feeds to uncover user co-sharing structures to discover political websites; then it uses a topic-agnostic classifier to score and rank newly discovered domains. To demonstrate the effectiveness of our system, we conduct an experimental evaluation in which we collect tweets related to the 2020 presidential impeachment process in the United States, and show that not only our system is able to discover new sites, but that a large percentage of these sites are indeed publishing fake news. We also design an integrated user interface to support fact-checkers and leverage their knowledge. Through this interface, fact-checkers can visualize domain interaction networks, query domain fakeness score, and tag incorrectly predicted results. Our proactive discovery system will expedite fact-checking process and can be a powerful weapon in the toolbox to combat misinformation.