I'm sorry for this silly question. When I use circ_vmpdf to calculate for the case kappa is big and data has close value with the mean, it return the value bigger than 1. For example:

circ_vmpdf(pi,pi,35)
ans =
2.3516

Did I mis-understand anything here?
I saw in the graphs of two functions based on kappa:

f1 = 1/(2*pi*besseli(0,kappa))
f2 = exp(kappa)

With high value of kappa, the increasing rate of second one is much higher than the first one so that it's not strange if the above case happened.