Two tools with clear verb-noun naming (list_supported_banks, get_bank_offers) and solid descriptions (118 and 156 chars respectively). Input schemas are present with proper JSON Schema structure, including regex pattern validation for bank_code. However, parameter descriptions are sparse, activity_status lacks explanation of what filtering by status accomplishes. Output schemas are not documented in the visible code, forcing LLMs to infer response structure. Error handling is present (UnsupportedBankError, BankSearchUnavailableError) but recovery guidance is minimal. Tool annotations (toolAnnotations=true) are declared but not visible in the schema definitions shown.

• Document the output schema for both tools. Example: 'list_supported_banks returns {banks: [{bank_code, bank_name, coverage_state, search_enabled, data_freshness}], timestamp}'. get_bank_offers should document the structure of returned offers (offer_id, offer_text, source_url, etc.).
• Add description to activity_status parameter: 'Filter results by offer activity status. Use "active" for current offers, "unknown" for offers with uncertain status. Omit to return all.'
• Enhance error messages with recovery hints. When returning unsupported_bank error, include: 'Call list_supported_banks to see available banks.'
• Explicitly set readOnlyHint=true in tool metadata for both tools to signal to agents that these are safe, non-destructive operations.
• If get_bank_offers can return large result sets, add a limit parameter (default 20, max 100) and document pagination strategy in the description.