Server has 10 tools with complete input schemas and descriptions visible in mcp.json. However, descriptions are generic and lack LLM-optimized guidance on WHEN to use each tool vs. alternatives. Parameter descriptions exist but are minimal (avg ~40 chars, below 72-char baseline). No output schemas documented. No error handling guidance. Tool names follow verb_noun convention well (generate_*, create_*, optimize_*, etc.), but lack dependency hints and recovery paths. One tool (create_project) is WRITE-class but descriptions don't emphasize irreversibility or confirmation patterns. Missing: pagination guidance, field-level chaining IDs, and structured error recovery.

• Document output schemas for all 10 tools. Specify: return type (object/array/string), required fields, field types, and examples. E.g., generate_component returns {html: string, jsx: string, explanation: string, accessibility_notes: string[]}.
• Enhance tool descriptions with WHEN-to-use guidance. E.g., 'Use generate_component to create a new component from scratch. Use get_shadcn_component if you want an existing shadcn/ui component. Use create_layout for multi-section page structures.'
• Add error handling guidance to descriptions. E.g., 'If Gemini API fails, the tool returns {error: string, retryable: boolean}. Retryable errors (rate limit, timeout) can be retried after a delay. Non-retryable errors (invalid input) require user correction.'
• Expand parameter descriptions to 50-100 chars with format/constraint guidance. E.g., 'description: string (50-500 chars, describe the component's purpose, layout, and key features)' instead of just 'Description of the component to generate'.
• Add confirmation pattern to create_project. E.g., add a 'dry_run' boolean parameter (default false) that returns the project structure without writing files, or require explicit 'confirm: true' parameter for irreversible operations.