메리츠증권 Open API 코드 생성 MCP. API를 찾아 바로 쓸 수 있는 Python 코드를 만들어 준다. 실제 호출은 하지 않는다.
Five tools with complete schemas and descriptions in Korean. All tools are READ_ONLY with proper annotations. Naming follows verb_noun pattern (list_, search_, get_, generate_, api_). Descriptions are detailed and context-rich (100-300 chars), explaining WHAT, WHEN, and HOW. Parameters have types and descriptions. However, output schemas are not formally documented in code, responses are inferred from dict returns. Error handling is minimal (get_api_detail returns error dict but no recovery guidance). No pagination despite search_api potentially returning many results. Tool composition is sound: search→detail→generate workflow is clear.
인증 절차·호출 규약·응답 코드·알려진 문제를 반환합니다.
api_type 을 호출하는 Python 코드를 만듭니다. REST 는 토큰 발급·요청·응답 판정까지, 실시간은 웹소켓 구독 코드를 만듭니다. 주문·환전은 confirm=True 전까지 전송하지 않는 안전장치가 들어갑니다. api_type='token' 이면 토큰 발급 코드만 만듭니다.
api_type 의 상세를 반환합니다 — 경로·파라미터·응답 필드·주의사항.
분류와 API 개수를 반환합니다.
자연어로 API 를 찾습니다. 예: search_api('삼성전자 현재가'), search_api('매수'), search_api('실시간 체결통보'). category 로 좁힐 수 있습니다 (account·domestic_market·overseas_market·trading·forex·reference·realtime·oauth2). 인증(접근토큰 발급·폐기)은 '토큰'·'인증'·'oauth'·'revoke' 로 찾습니다.
Output schemas not formally documented. Tools return dicts with inferred structure (e.g., get_api_detail returns {api_type, name, category, params, response_fields, ...} but no schema definition visible). LLMs cannot reliably extract fields or plan downstream calls without explicit output documentation.
search_api lacks pagination. No limit, offset, or page parameters visible. If CATALOG contains hundreds of APIs, search results could be massive, wasting tokens and degrading LLM reasoning. Baseline: paginated tools accept limit (default 20-50) and return total_count or next_cursor.
Error handling is minimal. get_api_detail returns {error, hint} on unknown api_type, but no recovery guidance for other failure modes (network, malformed input, etc.). Errors should categorize as retryable/user-fixable/fatal and suggest next steps.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 63 | 2026-07-28+ | v2 |
api_conventions tool has no input parameters and generic description ('인증 절차·호출 규약·응답 코드·알려진 문제를 반환합니다'). Unclear what structure it returns or when to call it. Should document output schema and explain when this is needed vs. get_api_detail.
Tool descriptions include Korean examples and instructions. While culturally appropriate for Meritz Securities domain, LLMs trained primarily on English may misinterpret or over-weight non-English text. Consider bilingual descriptions or English-first with Korean translations.