Two well-defined tools with complete schemas, clear descriptions, and proper tool annotations. ask_gemini has comprehensive parameter documentation with enums and constraints; get_gemini_consultant_config is minimal but correct. Both tools are read-only and properly annotated. Descriptions are concise (90 - 160 chars) and actionable. No critical gaps, but output schemas are not formally documented and error handling could be more granular.

• Document output schemas for both tools. Define the structure of ask_gemini's consultation result (e.g., reasoning, conclusion, confidence) and get_gemini_consultant_config's config object (model, thinkingLevel, etc.) as formal JSON Schema.
• Enhance error messages with recovery guidance. Instead of 'failed unexpectedly', return categorized errors: 'API rate limit exceeded, retry in 60s' or 'Invalid model ID, call get_gemini_consultant_config to see available models'.
• Add mutual-exclusivity validation for thinkingBudget and thinkingLevel. Document in both parameter descriptions: 'Cannot be used together with thinkingLevel' and validate in the handler, returning a clear error if both are provided.
• Specify numeric bounds in parameter descriptions. Update maxOutputTokens description to: 'Maximum output tokens (1 - 65536; defaults to model limit)' and thinkingBudget to: 'Token budget for thinking (0 - 32768; cannot be used with thinkingLevel)'.
• Return structured objects instead of JSON strings. Modify get_gemini_consultant_config to return a typed response object with fields like model, thinkingLevel, maxOutputTokens, etc., rather than stringified JSON.