It sounds like just a standardized set of api’s for interfacing with data sources. I’m not even sure why it seems to need a server in the diagrams from Anthropic. Couldn’t the data source just implement the API’s directly?
I am all for a standard set of API’s, I just don’t understand the middle man.
Not sure. To me it looks more like a structured description format - a tool card - rather than a ‘client/server’ protocol. Perhaps I’m missing something too ?
Afaik, we can just read this tool card, extract the important parts, and use a home-made tool-use instruction for the llm. So, at least we can use it as a library of llm-ready functions - until our llms just create them on the fly…