In example 9.4 of the Mplus User's Guide, there is a latent mediating variable y that is a random intercept. I know of literature where L2 independent variables are latent covariates, but is there any literature that specifically treats L2 latent covariates that are mediating variables?
I don't know that there is such a literature. Others? It may be a unique Mplus feature coming out of our general latent variable modeling framework. We put it in the UG because it was possible and seemed useful.
You may want to check the Preacher and Zyphur writings (Psych Methods) on multilevel mediation using Mplus - they may have touched on it.