Server exposes 24 tools with mostly complete schemas and descriptions. Naming follows verb_noun convention (fluent_*) consistently. Descriptions average ~80-120 chars, meeting baseline minimums. However, several tools lack visible implementation details (tools 3-8, 15 reference docs/assets/index-B2tCH_gi.js, a bundled asset file), making schema verification impossible for ~33% of tools. Parameter descriptions are present but often generic ('Name of the X'). No output schemas documented. Error handling and recovery guidance absent. STDIO transport caps protocol readiness at 50, limiting overall score despite decent definition quality.

• Migrate bundled asset tools (fluent_pbir_audit, fluent_pbir_apply_theme, etc.) from docs/assets/index-B2tCH_gi.js to explicit TypeScript source files in mcp/src/tools/ with visible schema definitions. This enables schema verification and raises inferred tool scores from 48 to 75+.
• Document output schemas for all tools. For lookup tools (fluent_tokens, fluent_components), specify: { items: [{name, description, category?, ...}], total, hasMore }. For generators (fluent_theme, fluent_code), specify the exact JSON/code structure returned.
• Add enum constraints to string parameters. E.g., fluent_powerbi_visuals.visualName should list valid visuals; fluent_design_guidance.topic should enumerate 'typography|color|spacing|motion|elevation|focus'. Replace free-form descriptions with formal constraints.
• Implement error handling and recovery guidance. For destructive tools (fluent_pbir_apply_theme, fluent_config set), add a dry_run parameter or return a confirmation prompt. For lookup failures, return available alternatives (e.g., 'Component not found. Did you mean: Button, Checkbox, Dropdown?').
• Add per-tool error classification in descriptions. E.g., 'Returns 400 if brandColor is invalid hex (user-fixable); 500 if theme generation fails (retryable); 403 if PBIR is read-only (fatal).' This guides agent recovery logic.