GHCi can evaluate individual IO actions, e.g.

ghci> writeFile "foo.txt" "bar"
ghci> readFile "foo.txt"
bar

but, as far as I know, this doesn't extend to other monads. Instead you have to compose an entire computation and then run it with the standard eliminators, e.g.

ghci> evalState 0 (put 1 >> get)
1

It would be quite nice if we could tell GHCi to temporarily jump into a different top-level monad, e.g.

-- we start in IO as usual
ghci> :monad runState 0
-- now we're in `State Int` with the initial state set to 0
ghci> put 1
-- we can evaluate individual State actions without the ceremony of `runState`,
-- and having to track the updated state ourselves between actions
ghci> get
1
ghci> :monad IO
-- now we're back in IO

I'm not sure how feasible this is (it seems tricky), but it would provide a very nice UX for GHCi in the context of larger Haskell applications.