Define an MCP tool named `{{tool_name}}`.

- `description`: what it does + WHEN an agent should call it (agents pick tools by description).
- `inputSchema` (JSON Schema): required params {{params}}, each with a `description`.
- Return: {{return_shape}}.

Rules:
- One tool = one capability. Don't overload.
- Make the description disambiguating vs. sibling tools.
- Validate inputs; return a clear error object on bad input, never throw raw.