We consider the issue of what should count as a resource for measurement-based quantum computation (MBQC). While a state that supports universal quantum computation clearly should be considered a resource, universality should not be necessary given the existence of interesting, but less computationally-powerful, classes of MBQCs. Here, we propose minimal criteria for a state to be considered a resource state for MBQC. Using these criteria, we explain why discord-free states cannot be resources for MBQC, contrary to recent claims [Hoban et al., arXiv:1304]. Independently of our criteria, we also show that the arguments of Hoban et al., if correct, would imply that Shor's algorithm (for example) can be implemented by measuring discord-free states.