The technical reason is that your degrees of freedom is the number of
clusters. If you have fixed effects, you consume all of these degrees
of freedom just by the fixed effects/country dummies, and you don't
have anything left to produce standard errors on your coefficients of
interest.