Ah, I understand. I thought it was the usual presented use case to apply an operation to a crypt directly and was confused since the title already stated otherwise.
Unfortunately, it's not that trivial. I tried a few offline wikis with varying success, but haven't had a ton of success using it day to day.
A trivial solution to this problem would be for the client to just download a copy of the entire Wikipedia database and pick out the particular article it's interested in. With a proper implementation of PIR, the server still needs to scan through the entire encrypted dataset (this is unavoidable, otherwise its I/O patterns would leak information) but the amount of information that needs to be sent to the client is much smaller.