concerning the question
is there a good reference for the construction of inductive datatypes (lists, trees etc.) in a topos with nno (assuming that my guess that such a construction is indeed possible is correct)?
The situation appears to me as follows. I those toposes where IZF (intuitionistic Zermelo Fraenkel set theory) is available (as e.g. Grothendieck and realizability toposes) business is as usual for functors preserving \omega-colimits: you simply construct mu(F) as colim_{n \in \omega} F^n(0). Notice, however, that there is a tacit appeal to the axiom of replacement when constructing the family (F^n(0))_{n\in\omega}. In general there is no reason why this family should exist. Of course, for the above mentioned examples like lists, trees etc. you can construct such inductive data types because both List(A) and Tree(A) appear as inductively defined subsets of P(\tilde{A}^N) where \tilde{A} is the partial map classifier. The reason is that trees and lists over A can be coded as partial maps from N to A. The point is that toposes (due to their impredicative nature) support inductive definitions of predicates on a type A given in advance. However, they don't support construction of inductively defined types as one cannot define sufficiently many families of types. It is not clear to me to which extent one can characterise those inductive types that can be reduced to inductively defined predicates in HAH. However, one knows that assuming W-types (`a la Martin-Loef) one can reduce most inductive types to W-types in extensional type theory. As far as I understand that's the reason why Moerdijk and Palmgren introduced lccc's with W-types as sort of ``predicative toposes''. Thomas