Organizations increasingly operate multiple Kubernetes clusters across data centers, cloud providers, and edge locations. This multi-cluster paradigm introduces a persistent inefficiency: resource fragmentation. At any moment, some clusters sit idle with surplus CPU and memory while others struggle under peak demand, yet no standardized, lightweight mechanism exists for sharing resources between clusters. Operators are forced to over-provision each cluster in isolation, leaving average utilization between 20% and 50%. This thesis designs, implements, and evaluates a multi-cluster Kubernetes resource sharing platform built around a centralized broker and lightweight per-cluster agents. Each agent collects its cluster's real-time CPU, memory, and GPU availability and publishes it to the broker, which maintains a federation-wide view of spare capacity.