usmart-mcp demonstrates solid tool definition quality with consistent naming, clear descriptions, and proper input schemas. All 9 tools follow verb_noun naming convention (health_check, get_account_summary, list_orders, etc.). Descriptions are concise (10-100 chars) and action-oriented. Input schemas use Zod validation with type constraints and descriptions. Tool annotations (readOnlyHint, destructiveHint, idempotentHint) are properly applied to read-only tools. However, output schemas are not explicitly documented in tool definitions, responses are returned as JSON without formal schema declarations. Error handling is minimal; tools return raw connector responses without recovery guidance. The server lacks pagination documentation in descriptions despite implementing it (e.g., get_positions accepts page/pageSize but description doesn't mention limits or total count).

• Add formal output schema documentation to all tools. Define response structure (e.g., 'Returns {health: boolean, timestamp: string}' for health_check) so LLMs can plan chained calls.
• Document pagination behavior in get_positions and list_orders descriptions: 'Returns a page of results (default 50 per page, max 100). Include total count in response to help agents determine if more pages exist.'
• Add error recovery guidance: wrap connector calls in try-catch, return structured errors like '{error: "symbol_not_found", suggestion: "Try get_quote with valid symbols from your account"}' instead of raw exceptions.
• Specify credential format constraints in save_usmart_configuration description: 'PEM-formatted keys must be valid RSA keys; passwords must be 8+ characters; phone number must be valid Singapore format.'
• Add input schema to configure_usmart and get_usmart_configuration_status (even if empty object) to clarify they accept no parameters and document their response structure.