Why the token count is the point
Tool schemas are sent on every request, whether or not a tool is called. A dozen tools with verbose descriptions and nested parameters routinely exceeds a thousand tokens of fixed overhead, and almost no framework shows you that number.
Writing them here means you see the cost as you add each one, which changes what you are willing to include.
The syntax
One tool per line: name, description and parameters separated by pipes. Parameters are comma-separated name:type pairs, and an exclamation mark marks one as required.
The output switches between Anthropic’s input_schema shape and OpenAI’s nested function shape. They are structurally different enough that hand-porting between them is a reliable source of bugs.