Proton-exchanged zeolites are effective Brønstedacid catalysts that exhibit molecular-level shape selectivity. Yet, the nonuniform, heterogeneous porous environment makes it challenging to understand and predict site-dependent reactivity that originates from the intricate, noncovalent interactions exerted by the catalyst pore walls, especially with large, flexible reactants. In this work, we developed a computational procedure consisting of generating quasi-transition-state (TS) complexes using force-fieldbased configurational-biased sampling and subsequent reaction path optimization using a density-functional theory-based nudgedelastic-band method. This approach allows us to capture how the TS configurations adapt to the local environment and to obtain a rough estimate of TS entropy via the number of accessible TS configurations at each active site. The resulting site-dependent TS energetics further enable the calculation of ensemble-averaged activation barriers. Using this approach, we studied the protolytic cracking of n-butane in TON, MFI, LTA, and FAU zeolites. It is found that in the MFI zeolite, while the tighter zig-zag and straight channels have lower TS energies, the more spacious intersection region potentially supports a much larger number of TS configurations. In addition, it is also found that if the reactant state is restricted to the vicinity of an active site, the computed site-specific barriers show a large variation across different zeolites and among different sites within the same zeolite, with values ranging from 182 to 218 kJ/mol. We argued that for kinetics-limited reactions, the reactant state should be taken to be the globally most stable configuration and showed that this treatment leads to ensemble-averaged barriers that are substantially more similar across zeolites, with the FAU and LTA zeolites having intrinsic barriers around 220 kJ/mol and the MFI and TON zeolites around 200 kJ/mol.