A system and method for cascading local conferences into a multi-site conference is disclosed. The local conferences are located in different sites where each site has endpoints served by a local multipoint control unit (MCU). An invitation to a multi-site conference is sent to participants at the different sites. The invitation includes a conference ID of the multi-site conference. The invited participants join the multi-site conference by calling a dial-in number of the local MCU associated with the caller's site. The caller gives the conference ID of the multi-site conference to the local MCU. The system searches the plurality of local MCUs to find those local conferences having the same conference ID conducted by different MCUs. For each found conference ID, the local MCUs are cascaded to establish a multi-site conference associated with the conference ID.