This server has critical gaps in definition quality. While 4 tools are explicitly registered with input schemas and descriptions, the descriptions lack depth and specificity required for LLM-driven tool selection. No output schemas are documented. Parameters are minimally described. The tool set conflates multiple concerns (ask_gemini doubles as a generic question answerer and code reviewer via prompt engineering). Error messages are generic ('Error calling Gemini'). The server_info tool is a fallback status check, not a production tool. Zero parameter descriptions exceed the 'basic' threshold of 72 chars. No dependency hints, no retry guidance, no actionable error recovery.

• Split ask_gemini into domain-specific tools: ask_gemini_qa (for general questions), answer_explanation (for clarification), and leverage existing gemini_code_review/brainstorm for those concerns.
• Expand tool descriptions to 150-250 chars. Example: ask_gemini → 'Ask Gemini to answer a question, explain a concept, or provide reasoning. Use when Claude alone lacks domain expertise. Returns free-form text; use gemini_code_review for code feedback instead.' This clarifies when to use each tool.
• Document output schema for each tool. Example: ask_gemini returns {type: 'object', properties: {response: {type: 'string'}, tokens_used: {type: 'number'}}, required: ['response']}. Enables LLM to structure downstream processing.
• Add parameter constraints in schema, not just description. Set temperature minimum: 0.0, maximum: 1.0. Add focus enum for gemini_code_review: ['security', 'performance', 'maintainability', 'general']. Enums let LLM pick valid options.
• Expand parameter descriptions to 60-100 chars with format/constraint details. Example: focus → 'Specific focus area for code review. Choose from: security (vulnerabilities, auth), performance (speed, efficiency), maintainability (clarity, structure), general (all aspects). Default: general.'
• Implement actionable error messages. Replace 'Error calling Gemini: {str(e)}' with categorized responses: 'API key invalid. Check GEMINI_API_KEY env var.' or 'Rate limited. Retry in 60 seconds.' or 'Prompt too long (>30k chars). Summarize or split input.'