Strong foundation with 11 well-named tools following verb_noun convention (kosis_status, kosis_search, kosis_data). All tools have descriptions (avg ~150 chars, within 10-1024 baseline). Input schemas present for all tools with type declarations and parameter descriptions. However, output schemas are NOT documented, responses are described in prose but lack formal JSON Schema definitions. Error handling is present (scrub() for secrets, _safe wrapper) but recovery guidance is minimal. Tool descriptions are domain-specific and helpful but could be more LLM-optimized (some exceed 200 chars with implementation details). Parameter descriptions are generally good but some lack explicit constraints (e.g., prd_se values not enumerated despite being fixed set Y/H/Q/M/D).

• Add formal JSON Schema definitions for all tool outputs. Document 'tables' as array of {tbl_id, org_id, tbl_nm, stat_nm, ...}, 'meta' as {total_count, pages, ...}, etc. This enables LLMs to plan chained calls.
• Convert prd_se, kind, format, and vw_cd parameters to enum types with explicit allowed values. Replace prose descriptions like '수록주기, Y(년)·H(반기)·Q(분기)·M(월)·D(일)' with enum: ['Y', 'H', 'Q', 'M', 'D'].
• Refactor kosis_data and kosis_collect descriptions to 100-150 chars. Move implementation details (auto-chunking, err codes) to a separate 'Advanced' section or docs. Focus description on: 'Fetch statistical data by table ID, period, and classification filters. Automatically handles large requests by splitting periods.'
• Enhance error responses with recovery guidance. Instead of {'error': 'err 20'}, return {'error': 'Invalid classification count', 'hint': 'Call kosis_meta(kind="OBJ") to see required axes', 'retryable': true}.
• Add per-parameter constraints in descriptions. E.g., max_records: 'Integer 1-1000 (default 30). KOSIS limits to 1000 per request.' This prevents LLMs from passing absurd values.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry