The most grug brain way of achieving this (also the least efficient) would be to ask for more data than you're actually interested in.

Another way is to publish the entire dataset and make it query-able from different sources.

Need to think if I can come up with a non-grug method.