When grammar-based techniques for natural language generation (and analysis alike) find their way into collaborative projects or actual application, big grammars tend to become hard to extend and debug. The MUG system represents a new tool set with a graphical debugging environment for functional unification grammars, which is designed to help grammar developers inspect the results of their work. The particular formalism supported is Multimodal Functional Unification Grammar, which is similar to Functional Unification Grammars, but supports several coordinated modes, such as voice prompts or structural and/or language-based screen displays. For each input description, the grammar can generate a range of coherent realization variants, which are ranked by a scoring function in order to optimize the output towards situational and device-related factors.