Great point. This is what we have been discussing internally. One option would be to let folks define an "object" (Read table structure) and display based on that.
I've kicked around some ideas for this problem a while back. I think letting users define a "view" like this is a pretty good approach.
I also think a reasonable default is to just pick the the N most popular (where "popular" means defined and non-empty) non-object/collection fields in the root document and display them in a table, then double-clicking or whatever on that row brings up the full document.
I'm admittedly not too tuned-in with the Mongo/Redis/Couch communities, but I do hear from a lot of folks using my native GUI http client tool to interact with their web APIs. I kind of assumed that most users of them were fairly happy to use low level/command-line tooling - is there a big unmet demand for native GUI tools for things like Mongo, Redis, etc?