Metal-organic frameworks (MOFs) are a class of porous materials constructed from metal or metal oxide building blocks connected by organic linkers. MOFs are highly tunable structures that can in theory be custom designed to meet the specific pore geometry and chemistry required for a given application such as methane storage or carbon capture. However, due to the sheer number of potential materials, identification of optimal MOF structures is a significant challenge. In this contribution we describe an automated technique for MOF design based on mathematical optimization. Optimization is performed on linkers represented by abstract space-filling shapes, in order to generalize the desirable geometric parameters describing linkers, and optimal shapes are projected to real molecules to illustrate realistic MOFs exhibiting the calculated properties. Six examined topologies of MOF and two distinct geometrical pore properties relevant to guest adsorption phenomena are explored. We demonstrate that the optimal shapes of linkers depend on both the topology and the property of interest and moreover that synthetically challenging linkers are not necessary to achieve the most promising candidate materials.