Two tools with clear naming (search_*, get_*) and reasonable descriptions (120 - 150 chars). Input schemas are visible and typed via Zod (SearchInput, GetBillInput). However, parameter descriptions are sparse or missing entirely, 'query' and 'identifier' lack detail on format/constraints. Output schema is undocumented; responses are raw JSON stringified without field-level guidance. No error handling guidance for agents (e.g., 'bill not found' recovery). Composition is sound (search → detail chain), but parameter constraints (e.g., limit 1 - 25) are enforced in code, not declared in schema descriptions for LLM visibility.

• Expand parameter descriptions: 'query' → 'Free text matched against bill title or identifier (e.g., "cannabis", "PC 456"). Partial matches supported.' 'identifier' → 'Bill identifier in format "PC 456" (project code + number) or "ps-123" (system ID). Case-insensitive.'
• Document output schema in tool descriptions. E.g., 'Returns array of {identifier, title, status, chamber, sponsors} objects.' Include field types and what each field represents.
• Add error recovery hints: 'If bill not found, try search_pr_legislation() with partial title or status filter to locate the measure.'
• Declare limit constraint in description: 'Max rows returned (1 - 25; default 20). Use pagination or multiple queries for larger result sets.'
• Return structured objects instead of JSON-stringified text. Modify tool handlers to return {content: [{type: 'text', text: ...}]} with a documented schema, or use tool annotations to hint at output structure.