Three tools with bilingual descriptions and clear regulatory focus, but significant gaps in schema completeness and parameter documentation. All tools have descriptions (10-150 chars), but input schemas lack parameter-level descriptions, critical for LLM parameter selection. No output schemas documented. Error handling is minimal. Tool names are action-oriented (ingredient_query, formula_review, compliance_check) but lack verb prefixes (search_, check_, validate_). Schema visibility is partial: formula_review and compliance_check have structured inputs with enums, but ingredient_query's input is minimal. No tool annotations (readOnlyHint, etc.) despite all being read-only operations.

• Add parameter descriptions to all tools. E.g., for ingredient_query: 'q: Chinese name (e.g., "水杨酸") or INCI name (e.g., "Salicylic Acid") to search across CN IECIC and EU CosIng databases.'
• Document output schemas for each tool. E.g., ingredient_query returns {query: string, cn_iecic_i: [{seq, chinese_name, inci_name, note}], cn_iecic_ii: [...], eu_cosing: [...], eu_annex_ii: [...]}.
• Rename tools with action verbs: ingredient_query → search_ingredient, formula_review → check_formula, compliance_check → validate_ingredient_compliance.
• Add readOnlyHint: true to all tool definitions in the MCP server registration to signal safety to agents.
• Enhance error messages with recovery guidance. E.g., 'DuckDB not found at {path}. Ensure data/bebd.duckdb exists and is readable.' or 'rules.json not loaded. Check data/rules.json exists and is valid JSON.'