Server has 11 tools with mostly complete schemas and descriptions in Korean. Naming follows verb_noun pattern (search_parts, get_product_detail, build_add). Descriptions are present but vary in quality (10-150 chars). All tools have input schemas with types and enums. However, output schemas are not formally documented, responses are formatted as text blocks rather than structured objects. Error handling exists but lacks recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite clear read/write distinction. Parameter descriptions are adequate but some lack format/constraint details.

• Convert text-formatted responses to structured JSON objects. E.g., search_parts should return {results: [{source: string, name: string, price: number, url: string}], total: number} instead of formatted text. This enables reliable field extraction for downstream tools.
• Add tool annotations: mark build_add and build_remove with destructiveHint=true; mark all read-only tools with readOnlyHint=true. This helps agents reason about side effects.
• Enhance error messages with recovery guidance. E.g., 'Danawa blocked (HTTP 429). Retry in 60 seconds or call proxy_status() to check service availability.'
• Document parameter formats in descriptions. E.g., productCode: 'Numeric product code from Danawa URL (e.g., 12345678)'. Specify typical ranges for limit (e.g., 'Recommended: 10-20 for search, 5-10 for price comparison').
• Add optional build_id parameter to build_* tools to support multiple concurrent builds. Return build_id in build_status response so agents can reference specific builds in subsequent calls.