Policy negotiations of occur behind closed doors. While minutes from those meetings exist, it typically takes a certain period, e.g., 25 years, until those minutes are released. Political scientists, however, are keen to explore data close to the present for their research questions. We explore Multi-Agent Systems built from Large Language Models (LLM MAS) to model possible negotiations at the top level. The task the agents solve is to get from an initial negotiation document A through several rounds of negotiation and voting to a final outcome document B. Empirically, we simulate negotiations from the committees in the European Parliaments. Given the highly detailed documentation through live streams and position briefings, and the institutionalised course of negotiations, these committee negotations offer ample empirical foundation to fine-tune and validate the models. Our method finds ample application for all keen to simulate and study possible negotiation behaviour, not only in Political Science, but in law, business, international relations or psychology.