Semrush MCP Pack provides 6 well-structured SEO analytics tools with complete input schemas and clear descriptions. All tools follow verb_noun naming (domain_overview, domain_organic_keywords, etc.) and include detailed parameter documentation. However, output schemas are not formally documented, descriptions mention CSV column abbreviations (Ph, Po, Nq, etc.) but the actual JSON response structure is not declared. Error handling is basic (throws on non-2xx or 'ERROR' prefix) with no recovery guidance. No tool annotations (readOnlyHint, destructiveHint) despite all being read-only. Parameter descriptions are strong (72-char average baseline met), but missing enum constraints for database codes and target_type values.

• Add formal output schema documentation to each tool. Example: 'Returns array of objects with fields: {Ph: string (keyword phrase), Po: number (position), Nq: number (search volume), Cp: number (CPC), ...}'. Declare whether results are paginated and if there is a total count.
• Convert database parameter descriptions to enum constraints in inputSchema: {type: 'string', enum: ['us', 'uk', 'de', 'fr', 'es', 'ca', 'au'], description: '2-letter Semrush regional database. Default: us.'}
• Add readOnlyHint: true annotation to all 6 tools in their definitions to signal to agents these are safe, idempotent, read-only operations.
• Enhance error messages with recovery guidance. Catch Semrush error codes (120=bad key, 40=empty result, etc.) and return: {error: 'Invalid API key', suggestion: 'Verify _apiKey and Semrush account status', retryable: false}
• Document pagination behavior. If Semrush API supports offset/limit, expose display_limit in all tools and clarify: 'Returns up to display_limit rows. For large result sets, call multiple times with different offsets.' If no pagination, state: 'Returns all matching results (capped at Semrush API limit).'