Server has 5 tools with clear verb-noun naming (get_*, list_) and reasonable descriptions (avg 150 chars). However, critical gaps undermine production readiness: (1) Most tools lack input parameter descriptions, get_page_overview, get_design_tokens, list_components have empty required arrays but no guidance on what they return or when to call them; (2) get_component_detail and get_node_css accept node_id OR node_name but lack documentation of mutual exclusivity or fallback behavior; (3) No output schemas documented, LLMs cannot plan downstream calls or extract structured data; (4) No error handling guidance, 'Node not found' responses lack recovery hints (e.g., 'call list_components first'); (5) include_children defaults to true but no rationale or consequence documented. Tool names are clear and follow verb_noun convention, but parameter design and output documentation are weak.

• Add explicit output schema documentation for each tool. Example for get_page_overview: 'Returns {frames: [{id, name, children}], designTokens: {colors, fonts, spacings}, summary: string}'.
• Document parameter relationships: 'Either node_id (exact match, e.g. "1:23") or node_name (partial match) required; if both provided, node_id takes precedence.'
• Enhance error messages with recovery hints: 'Node not found. Call list_components to see available nodes, or search by partial name with get_component_detail(node_name="...").'
• Add discovery guidance to tool descriptions: 'Call get_page_overview first to understand structure, then list_components to find specific nodes, then get_component_detail for full styles.'
• Document include_children behavior: 'Set to false to reduce output size for large component trees; default true includes all descendants for complete CSS generation.'