This project addresses resource allocation in multi-agent planning under uncertainty, where multiple agents make independent decisions to achieve their own goals while sharing finite resources. In these settings, stochastic environments leave agents uncertain about their own future resource, which makes coordination and allocation a challenge. Prior work on resource-constrained planning under uncertainty typically enforces constraints in expectation or with a fixed probability of violation, but such formulations ignore the distribution of outcomes — in particular, how severe worst-case resource overuse can be. This project instead formulates and solves risk-constrained shared-resource planning problems, where agents maximize reward subject to a constraint on a distortion risk measure — such as Value-at-Risk, Conditional Value-at-Risk (CVaR) — applied to shared resource usage. We plan to investigate these challenges in the context of Multi-Agent Markov Decision Processes (MMDPs), across both cooperative and non-cooperative settings. This includes exploring model-based multi-agent reinforcement methods for allocating resources under risk constraints when agents may act strategically, as well as developing methods that account for how individual agents contribute to overall risk in cooperative settings. We further intend to relax common independence assumptions between agents by considering how worst-case dependencies between agents affect risk-constrained planning, with the aim of developing methods that remain valid under more general and realistic dependence structures. Throughout, we plan to experimentally evaluate our methods against state-of-the-art techniques on synthetic multi-agent domains. Realizing these contributions computationally requires large numbers of independent simulation runs to accurately resolve tail-risk estimates via rejection sampling, particularly as the number of agent grows, alongside model checking of large state spaces (using PRISM) to validate our planning methods against ground-truth guarantees where feasible. This computational demand is the basis of the resource request detailed below.