I think there may be a misunderstanding here - mal is not the language you make the lisp (or any other language) in, it is the language you are implementing. Indeed, there is an implementation in Racket, and even in mal itself.

The point of mal is to guide an implementer through the steps necessary for making an interpreter using a language they are comfortable in. It gives some insight into how various constructs are implemented, such as closures and tail call optimization. The purpose is not so much to make a lisp-like language (although that is the outcome), but to learn what goes into making an interpreter for one - especially when you can't lean on the constructs that already exist in the host language.