This server has clear tool naming conventions and comprehensive parameter schemas. All tools have descriptions and input schemas with type definitions. However, output schemas are not documented, error handling lacks recovery guidance, and tool descriptions could be more specific about when and why to use each tool. The server follows good naming patterns (verb_noun: generate_text, analyze_image, list_models, code_review) and includes tool annotations. Parameter descriptions are present but some lack clarity on constraints and dependencies.

• Document output schemas for all tools. Specify what fields each tool returns (e.g., generate_text returns {content: string, tokensUsed?: number}, list_models returns {models: [{id: string, displayName: string, capabilities: string[]}], total: number}).
• Add structured error responses with categories. Instead of fail(message), return {ok: false, error: {code: 'USER_NOT_FOUND'|'INVALID_PARAM'|'RATE_LIMIT'|'API_ERROR', message: '...', recovery: 'Try X next'}}.
• Expand tool descriptions to 50-200 characters with explicit WHEN and WHY guidance. E.g., generate_text: 'Generate text using Google Gemini. Use for creative writing, summarization, Q&A, or code generation. Returns plain text or JSON if jsonMode is enabled.'
• Add oneOf constraint in analyze_image schema to enforce mutual exclusivity: either imageUrl OR imageBase64, not both.
• Document parameter dependencies and constraints in descriptions. E.g., 'conversationId (string, optional): UUID for multi-turn conversations. If provided, prior turns are retrieved from server cache; omit to start a new conversation.'
• Expand list_models description to explain its discovery role: 'List all available Gemini models. Call this first to check supported model names and capabilities before creating generate_text or code_review requests.'
• Add result/pagination guidance if applicable. If tool responses can be large, document any internal limits (e.g., 'Top 20 models returned'; 'Context limited to 4000 tokens').
• Consider adding per-tool error recovery examples in the server's instructions or README, linked from tool descriptions.