Managing risks from extreme events will be a crucial component of climate change adaptation. In this study, we demonstrate an approach to assess future risks and quantify the benefits of adaptation options at a city-scale, with application to flood risk in Mumbai. In 2005, Mumbai experienced unprecedented flooding, causing direct economic damages estimated at almost two billion USD and 500 fatalities. Our findings suggest that by the 2080s, in a SRES A2 scenario, an 'upper bound' climate scenario could see the likelihood of a 2005-like event more than double. We estimate that total losses (direct plus indirect) associated with a 1-in-100 year event could triple compared with current situation (to 690 - 1890 million USD), due to climate change alone. Continued rapid urbanisation could further increase the risk level. Moreover, a survey on the consequences of the 2005 floods on the marginalized population reveals the special vulnerability of the poorest, which is not apparent when looking only through a window of quantitative analysis and aggregate figures. For instance, the survey suggests that total losses to the marginalized population from the 2005 floods could lie around 250 million, which represents a limited share of total losses but a large shock for poor households. The analysis also demonstrates that adaptation could significantly reduce future losses; for example, estimates suggest that by improving the drainage system in Mumbai, losses associated with a 1-in-100 year flood event today could be reduced by as much as 70%. We show that assessing the indirect costs of extreme events is an important component of an adaptation assessment, both in ensuring the analysis captures the full economic benefits of adaptation and also identifying options that can help to manage indirect risks of disasters. For example, we show that by extending insurance to 100% penetration, the indirect effects of flooding could be almost halved. As shown by the survey, the marginalized population has little access to financial support in disaster aftermaths, and targeting this population could make the benefits of such measures even larger. While this study explores only the upper-bound climate scenario and is insufficient to design an adaptation strategy, it does demonstrate the value of risk-assessment as an important quantitative tool in developing city-scale adaptation strategies. We conclude with a discussion of sources of uncertainty, and of risk-based tools that could be linked with decision-making approaches to inform adaptation plans that are robust to climate change.