The tools right now act as an API wrapper and just dump the API output to the LLM. For the internal tools we should parse the output and return only relevant fields in some friendly markdown format.
Markdown may no be needed or relevant, we need to test first but we need to return only relevant fields (if dataset view definition is available)