Leave the tests out of the release. Build the release, then start it from a CT run (test_server has a nice way to start slave nodes). Now you can call into the other node to execute test cases.
I find that this method is often easier to get working.