It is not implemented. CDF of MVN doesn't have a clearly obvious definition. Usually people mean the total mass in a region whose points are elementwise less than a given point. But this integral is very difficult in general (and IMHO not actually a very useful piece of information). Approximate methods exist, at least in small dimensionalities, but we never implemented them in tfp. Usually the right answer here is to step back and find a "better" question to ask about the distribution than the usual CDF. Sorry! But hope this lends some clarity as far as tfp is concerned.