I have never done this myself, but I heard my friends employ R for multi-level modeling.

Have you ever tried the SEM builder? It worked for me. Others, you may try Mplus.

Yes, there are numerous (free) packages to do multilevel modeling with covariates in R. The popular frequentist ones are nlme and lme4. For the Bayesians, brms and rstanarm are both quite nice.

Have a look at 

Steele, Fiona, Vignoles, Anna and Jenkins, Andrew (2007) The effect of school resources on pupil attainment: a multilevel simultaneous equation modelling approach. Journal of the Royal Statistical Society, Series A, 170 (3). pp. 801-824. ISSN 0964-1998

http://eprints.lse.ac.uk/26481/

they use a set of  instruments in a multilevel model to allow the fitting of a simultaneous relation between two response (educational attainment and expenditure) in a multilevel model

Thank you all for your responses. 

Kelvyn, I will see whether I can find an implementation.

Bram, do you have a reference for that? What happens to the SEs, I assume bootstrapping is required, anything else?

Hey, did you ever find an acceptable approach that works in STATA? I have done something where I bootstrap the second stage model. But, I would really like to do a simultaneous model. I am not sure if there is anything out there.

How about running the first stage regression and adding the predicted values of endogenous variables in the second stage regression, which can be estimated by XTMIXED command?

Please I want your help. I have micro (individuel level) and macro (country level) data which I will estimate a multilevel simultaneous equation. Why I can do it in stata

Have a look at

http://www.applied-ml.com/product/multiprocess.html

Can we use 2SLS to instrument a higher-level variable? At the initial stage, does it make sense to simply regress the observed higher-level variable on its IV and the other covariates at both lower- and higher- levels? @Bram Hogendoorn

Zequn: No, it doesn't I am afraid. Youd need to estimate a higher lever first stage