There is a full and faithful embedding of the category of schemes into the $2$-category of (edit: stricly Henselian ringed) topoi, which sends each scheme $X$ to the topos $Sh\left(X_{et}\right)$ (in fact, this is fully faithful even on Delign-Mumford stacks, see: https://eudml.org/doc/90454 Theorem 50, and Definition 43). So, given a geometric morphism $Sh\left(X_{et}\right) \to Sh\left(Y_{et}\right)$ respecting structures sheaves is the same as given a morphism of schemes $X \to Y$. In particular, any functor $X_{et} \to Y_{et}$ which induces such a geometric morphism (edit: respecting structure sheaves), also induces a morphism of schemes, but not every geometric morphism can be realized as a morphism of sites.
Now, the structure sheaves are realized as a geometric morphism $Sh\left(X_{et}\right) \to \mathscr{T},$ with $\mathscr{T}$ the topos of sheaves on the site $Ftyp$ of all schemes of finite type, with the etale topology. This is the classifying topos for strictly Henselian rings. In fact, we can enlarge $Ftyp$ to all schemes locally of finite type $LFtyp$ without changing the topos. If $X$ is also locally of finite type, there is a canonical functor $\pi_X:X_{et} \to LFtyp$ sending each etale map $X' \to X$ to $X'.$ This is a morphism of sites which induces the geometric morphism $Sh(X_{et}) \to \mathscr{T}$ encoding the structures sheaf. In particular, any map of sites $f:X_{et} \to Y_{et}$ commuting over their projections to $LFtyp$ encodes a morphism $X \to Y.$ However, this is too strict, since maps of ringed topoi need not respect structure sheaves up to isomorphism, but only induced a map (like in ringed spaces), i.e. it suffices to have a natural transformation $\alpha:\pi_Y \circ f \to \pi_X$.