Material Icons server has well-structured tool definitions with consistent naming (verb_noun pattern), comprehensive parameter schemas using Zod validation, and clear descriptions. However, output schemas are entirely undocumented, responses are wrapped in generic text/error format with no structured field definitions. Error handling is minimal (generic fail() wrapper). Tool descriptions average ~120 chars (within baseline 194), but lack dependency hints and recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite clear risk stratification. Parameter descriptions are present but terse. Missing: output schema documentation, error classification, confirmation patterns for destructive ops, and LLM-optimized recovery guidance.

• Document output schemas for all tools. E.g., search_icons should return {results: [{name, style, url}], total_count}. Include field types and descriptions so LLMs can extract and chain data.
• Implement error classification. Wrap errors with {error_type: 'retryable'|'user_fixable'|'fatal', message, suggestion}. E.g., 'Icon not found. Try search_icons() first.' for user_fixable.
• Add tool annotations: readOnlyHint=true for search_icons, audit_icons, detect_project_environment, validate_icon_usage; destructiveHint=true for delete_icon, update_icon; idempotentHint=true for import_icon (never overwrites).
• Enhance parameter descriptions with dependency hints. E.g., import_icon: 'registerFlutterAsset only applies when target=flutter. Call detect_project_environment first to confirm Flutter project.'
• Add dry-run support to destructive tools. E.g., delete_icon with dry_run=true returns what would be deleted without executing. Enables safer agent planning.