Marcus Gunn pupil describes the finding during the swinging-flashlight test whereupon the patient's pupils appear to dilate instead of constrict when the light swings from the unaffected "good" eye to the affected "bad" eye.
It is important to remember that there is no anisocoria in this case. The pupils remain the same size as each other at all times.
Upon shining the light into the "good" eye, both pupils will constrict. However, when the light moves to the "bad" eye, the full strength of the light will not be perceived and both pupils will appear to dilate.
The Marcus Gunn phenomenon is a relative afferent pupillary defect. That is to say, the "bad eye" can still perceive light and respond to it, but not as much as the "good eye"; the bad eye is relatively less responsive than the good eye, but both eyes are still responsive to light. If you shine the light in the bad eye, both pupils will constrict (due to the still-intact consensual light response). However, if you shine the light in the "good eye", the pupils will constrict even more. It is as if you are shining a light of lesser intensity at the bad eye.
In context of the swinging flashlight test, you first shine the light in the good eye, causing full pupillary contraction in both eyes. Then you move the light to the bad eye. The bad eye perceives this same light as if were not as bright, and thus causes the pupils to constrict less. This gives the illusion that both pupils are now dilating as a response to the light. They are actually still constricting in response to the light, but constricting less than when the light was shining at the good eye, because the bad eye perceives a dimmer light. But relative to the previous maximal dilation from shining the light at the good eye, the pupils now dilate. Had you started with the light shining on the bad eye first, you'd see both eyes constrict slightly. This distinguishes the Marcus Gunn Pupil from a total CN II lesion, in which the bad eye perceives no light. In that case, shining the light at the bad eye produces no effect. In any case the patient themselves should report that they are totally blind in the unreactive eye.