We have done an installation of Real Time Speech.
After we have finished the installation we can see the calls are routed correctly to the RTS snap-in. Only the Media Server is not anchored into the call. Through attributes we tell that all calls should be analyzed.
Should this be working or should we create web services to start a search on the call and make sure the Media Server is anchored into the call?

Default behavior is to only bring the media server into the media stream when a speech search is in progress. It can be "shuffled" in mid-call if needed so as to conserve media resources.

Are you observing that the media server is not in the media stream even when a speech search is active? I believe that there is also a setting to always include the media server through the duration of the call to avoide any media clippping when shuffling. Based on your initial post, I think you may not have a speech search in progress so you are likely encountering expected behavior.