If all else fails, you can try an approach like with vnccapture (except you'll need to write rdccapture), together with some "image recognition" that recognizes bitmaps and translates them back to UI elements again. This is likely highly brittle but might still work.