Strong tool design with comprehensive Korean financial disclosure APIs. All 18 tools have clear names starting with action verbs (get_, search_, resolve_), well-structured input schemas with proper type definitions, and Korean-language descriptions that convey purpose. Output schemas are implicitly documented via structured tool handlers. However, descriptions are consistently in Korean (limiting LLM-agnostic clarity), several tools have high-complexity parameter surfaces with undocumented dependencies (e.g., search_disclosures combines preset + date range logic), and no visible error handling guidance for common failure modes (API rate limits, invalid corp codes, malformed XBRL). The dart_raw passthrough tool (18) intentionally exposes raw API parameters with minimal validation, trading safety for coverage. Overall, the set demonstrates solid engineering discipline but falls short of production A-grade due to missing error recovery patterns and implicit output schemas.

• Translate all tool descriptions and parameter annotations to English (or provide bilingual descriptions). Monolingual Korean descriptions lock the server to Korean-speaking LLMs only.
• Document the full output schema for each tool. Create a schema.ts or similar that defines the shape of every response (e.g., CorpCodeResult = { code: string; name: string; ... }). Publish these in README or OpenAPI spec.
• Add explicit error handling for common failure modes: (1) corp code not found → suggest search_disclosures or resolve_corp_code; (2) DART rate limit → retry with backoff; (3) XBRL parse error → offer fallback to raw document or raw format. Return structured errors with actionable guidance.
• Refactor search_disclosures: split preset-driven search from date-range search. Create search_disclosures_preset(corp, preset, days?) and search_disclosures_by_date(corp, begin, end, size, limit, all_pages) to eliminate parameter interdependencies.
• Add explicit max limits to pagination parameters. E.g., 'limit (1-500, default 50)', 'size (1-100, default 20)' in parameter descriptions. Validate inputs and return errors if exceeded.
• Rename buffett_quality_snapshot to clarify mode selection. Either split into buffett_quality_snapshot(corp) and buffett_quality_compare(corps) OR add explicit mode parameter: buffett_quality(corps, mode: 'snapshot'|'compare'). Document in description.
• Consolidate governance queries. Consider merging get_shareholders, get_executive_compensation, get_major_holdings into a single get_governance_summary(corp, year, sections: ['shareholders', 'compensation', 'holdings']) to reduce ambiguity and tool count.