There is indeed a confusing presentation of information. If you examine the bottom of the detail page for mp-18717, you’ll notice we have done a GGA+U calculation for the structure optimization, but we have not yet run a GGA+U calculation for the band structure.
Because the optimized structure is via a GGA+U calculation, it is a “GGA+U material”, and unless there is a GGA+U band structure calculation, the band gap value displayed is based on the loose k-point mesh of the structure optimization. If there was a GGA+U band structure calculation, the band gap value would be more accurate. The API (e.g. via MPRester) reports no bandstructure because there isn’t a GGA+U one corresponding to the structure optimization. In contrast, the website shows any available bandstructure, even if it isn’t GGA+U. This explains the inconsistency between the displayed bandstructure and the band gap field on the detail page in this case.
The behavior of the website and API should be consistent. I think it is better to not present a GGA bandstructure on the website in this case, i.e. I think the API behavior is correct. The MP team will discuss how best to handle this. Thank you for reporting the issue.