The explainability of machine learning black-box models is key for designing and adopting AI technologies by end users. XAI tools such as SHAP or LIME have been purposely developed to support such explainability but their exploration in the HCI community has been limited. This paper reports a systematic review of 142 papers targeting design, use or evaluation of XAI tools with the aim to investigate their different types, users, application domains, input and output data sets, and their user interfaces. Findings indicate a broad range of XAI tools but extensive use of a few, a prevalence of AI experts as users rather than evaluators of these tools. We discuss our findings arguing for the need to move beyond the design of novel XAI tools towards increasing their use and comparative evaluation. We also argue for the need for HCI-grounded user interface design for XAI tools and advance an initial design space for AI-HCI research integrating AI affordances with the application domains of XAI tools mapped to key HCI research areas.