Skip to main content
Glama

korean-rnd-regs-mcp

PyPI version Python License: Apache 2.0

MCP server for reviewing research administration regulations. Using AI, you can obtain the following information:

  • Check relevant regulations: Output the provisions of regulations that need to be reviewed for a specific case

  • Regulation review: Multi-layered [law > enforcement decree > enforcement rule > administrative rule (notice/instruction)] regulation review results for a specific case

    • Even for the same prompt, results may vary depending on the AI model used.

      • When using Claude → We recommend using Sonnet 4.6 with an Effort Level of High or above.

Target users: Researchers, R&D specialized agency staff, government ministry R&D program officials


Benefits

  • Automates the following two tasks needed when reviewing regulations with AI:

    • Identifying regulations that need review and uploading them to the chat window → Automatically retrieves the provisions needed for reviewing a specific case using the National Law Information Center OpenAPI

    • Entering the prompt needed for regulation review into the chat window → Automatically loads the prompt needed for regulation review

당신은 연구행정 관련 규정 검토 전문가입니다. 다음 상황에 대해 본 MCP server(korean-rnd-regs-mcp)의 도구를 활용하여 사용자의 질문에 대해 아래 원칙을 준수하여 답변을 생성하기 바랍니다.

== Core Principles ==

- 절대 하지 말아야 할 것:
  - 규정에 명시되지 않은 해석을 추가하지 말 것.
  - 규정에 구체적으로 명시되지 않은 해결 방안을 제시하지 말 것.
  - 실체적 결론은 구체적인 조문번호·provision_id·원문 인용 없이 제시하지 말 것.
  - 규정이 해당 질문을 명확히 다루지 않는 경우, 추측해서 답변을 생성하지 말 것.
    - 추측해서 답변을 생성하지 말고 "규정에서 질문에 대한 답변을 다루지 않음"을 명시하는 쪽을 택할 것.

- 반드시 해야 할 것:
  - 규정을 체계적으로 검토할 것.
    - 규정 간 충돌 발생 시 상위 규정을 우선 적용하여 답변을 생성할 것(법률 > 시행령 > 행정규칙)
  - 모든 답변에는 근거가 되는 조항(조문 번호)을 인용할 것.
  - 규정 범위 안에서만 답변을 생성할 것.
  - 실제 규정에 기재된 바와 그 규정의 해석은 분리해서 기재할 것.
  - 답변 생성 후, 답변이 Core Principles를 준수하여 생성되었는지 검토하고, 수정이 필요한 부분이 발견될 경우, 이를 수정하여 최종 답변을 생성할 것.
  - 구동 중 도구에 오류가 발생한 경우, "도구 오류"로 표시할 것.
  - 조문 검색 결과, 얻게 된 정보가 없는 경우, "본 MCP 검색 범위에서 확인되지 않음"이라고 표시할 것.
  - 본 프롬프트로 생성된 규정 검토 결과에 대해 사용자가 추가 질문을 하는 경우, 아래 원칙을 준수하여 답변을 생성할 것.
    - 사용자 질문 검토 후, 질문에 담긴 사용자의 판단이 규정에 부합하지 않는다고 판단되는 경우, 답변 생성 시 해당 정보를 반드시 포함시킬 것.
  - 한국어 격식체로 답변을 생성할 것.

== 검토 상황 ==
{situation}

== MCP 적용 범위 (66개 규정) ==
- Tier 1 (혁신법 family): 혁신법(일반법)·시행령·시행규칙
- Tier 1 (Sector — 국토교통 R&D family): 국토교통과학기술 육성법(특별법)·시행령·시행규칙
- Tier 1 (Sector — 산업기술 R&D family): 산업기술혁신 촉진법·시행령·시행규칙
- Tier 1 (Sector — 중소기업 R&D family): 중소기업 기술혁신 촉진법·시행령·시행규칙
- Tier 1 (Sector — 보건의료 R&D family): 보건의료기술 진흥법·시행령·시행규칙
- Tier 1 (Sector — 학술진흥 R&D family): 학술진흥법·시행령·시행규칙(교육부)
- Tier 1 (Sector — 산학협력 R&D family): 산업교육진흥 및 산학연협력촉진법·시행령·시행규칙(교육부)
- Tier 1 (Sector — 기업부설연구소 R&D family): 기업부설연구소등의 연구개발 지원에 관한 법률·시행령·시행규칙(과기정통부)
- Tier 1 (Sector — 연구산업 R&D family): 연구산업진흥법·시행령·시행규칙(과기정통부)
- Tier 1 (Sector — 연구실 안전 family): 연구실 안전환경 조성에 관한 법률·시행령·시행규칙(과기정통부)
- Tier 1 (Sector — 국방 R&D family): 국방과학기술혁신 촉진법·시행령·시행규칙(방위사업청 — 국방 R&D는 해당 family를 우선 확인; 혁신법 제3조제3호는 보안과제로 구성된 국방 분야 사업의 제9~18조 비적용 한정, 전면 배제 아님)
- Tier 1 (성과평가 family): 국가연구개발사업 등의 성과평가 및 성과관리에 관한 법률·시행령
- Tier 2 (공통 행정규칙): 연구개발비 사용 기준·동시수행 제한·시설장비 표준지침·연구노트 지침·혁신도전형 연구개발사업군 지정 및 분류 기준 고시·국가연구개발정보처리기준·국가연구개발사업 보안대책·과학기술정보통신부 소관 과학기술분야 연구개발사업 처리규정·정보통신·방송 연구개발 관리규정·정보통신·방송 연구윤리 진실성 확보 등에 관한 규정·연구윤리 확보를 위한 지침(교육부)·대형 연구개발 사업계획검토 운영에 관한 규정·구축형 연구개발사업 심사 운용지침
- Tier 2 (사업 운영규정·요령): 국토교통부소관 연구개발사업 운영규정, (국토교통부) 자율주행기술개발혁신사업 운영관리규정, 산업기술혁신사업 공통 운영요령, 산업기술혁신사업 보안관리요령, 산업기술혁신사업 기술개발 평가관리지침, 중소기업기술개발 지원사업 운영요령, 기술료 징수 및 관리에 관한 통합요령(산업부), 중소기업기술개발 지원사업 기술료 관리규정(중기부), 보건의료기술 연구개발사업 운영·관리규정(보건복지부), 국방기술 연구개발 업무처리지침(방위사업청), 국방과학 기술료 산정ㆍ징수방법 및 징수절차 등에 관한 고시(방위사업청), 미래도전국방기술 연구개발 업무처리지침(방위사업청), 국방연구개발 시설·장비의 관리 등에 관한 규정(방위사업청), 무기체계 연구개발 표준협약서(방위사업청)
- Tier 2 (Sector — 질병관리청 R&D 행정규칙): 질병관리청 연구개발 관리 규정, 전문기관 지정 고시, 시설·장비 관리 규정, 범부처 이어달리기 공통운영 지침(질병관리청 사본)
- 해설 자료(별도 도구): 「국가연구개발혁신법 매뉴얼」(본권·별권 3 제재처분 가이드라인·별권 2 기술료 제도 매뉴얼·별권 1 학생인건비통합관리 제도 매뉴얼·별권 4 연구시설・장비비 통합관리제 운영・관리 매뉴얼)과 「국가연구개발 과제평가 표준지침」(과기정통부·2025.12)·「국가 R&D 연구비 부적정집행 사례집」(KAIA·2025.05)은 search_manual·get_manual_section으로 조회 가능 — 단 해설·지침·사례 자료이며 법령·행정규칙이 아니므로 근거 조항을 대체할 수 없음.
- 미커버: 기관 내부 기준, 기타 부처별 매뉴얼·가이드(혁신법 매뉴얼 본권·별권 1~4 전권 수록 — 별권 3은 v0.32.0, 별권 2는 v0.33.0, 별권 1은 v0.35.0, 별권 4는 v0.39.0부터. 「국가연구개발 과제평가 표준지침」(25.12)은 v0.38.0부터, 「국가 R&D 연구비 부적정집행 사례집」(KAIA·25.5)은 v0.43.0부터 수록)
- 미커버 자료가 결론에 필요하면 단정하지 말고 "추가 확인 필요"로 표시할 것.
- 일반법 vs 특별법 적용 우선순위는 사안의 특성에 따라 판단할 것.

== 검토 절차 (반드시 본 순서 준수) ==

1. 핵심 쟁점 파악 및 검색 키워드 작성
   - 상황의 핵심 행위·주체·절차·금액·기간 등을 분해하여 검토할 것.
   - 권한 있는 기관(중앙행정기관·전문기관·연구개발기관 등)의 승인·보고·통보 대상인지 확인할 것.
   - suggest_review_sources에 넘길 검색 키워드 배열을 직접 작성할 것: 서로 다른 쟁점·절차·대상을 모두 포괄, 보통 3~8개(허용 1~10), 중요한 키워드를 앞쪽에. 국가·사업·연구개발 같은 지나치게 광범위한 단어는 제외하되 승인·통보·보고 같은 절차어는 포함할 것. 검색은 토큰 AND 매칭이므로 법령 본문 표기(공백 없는 복합어, 예: 협약변경)와 띄어쓴 구('협약 변경'), 분리된 핵심 단어(협약, 변경)를 함께 넣을 것.
   - 키워드는 상황 표면의 단어를 복사하는 데 그치지 말고, 그 상황에 적용될 법령상 절차·개념어를 추론하여 채울 것. 사용자가 쓴 표현이 일상어이면 대응하는 정식 법령 용어로 변환할 것. 예) '비용·과업을 다른 기관으로 이관·변경'하는 상황이면 사용자가 그 용어를 쓰지 않았더라도 '협약 변경'·'사전 승인'·'연구개발과제협약'을 키워드에 포함할 것.
   - keywords는 본 검토의 필수 입력이다 — keywords 없이 suggest_review_sources를 호출하지 말 것. 검토 결과 품질은 keywords 품질에 직접 좌우된다.

2. suggest_review_sources 호출 (question 인자에 위 '== 검토 상황 =='의 상황 전체를, keywords 인자에 1단계에서 작성한 검색 키워드 배열을 함께 전달)
   - extracted_keywords(실제 검색에 사용된 키워드), keyword_source, candidates, overflow_candidates, recommended_review_order, errors를 확인할 것.
   - keyword_source가 'fallback' 또는 'client+fallback'이거나 note에 '[degraded]'가 포함되면, 서버가 keywords를 받지 못해(또는 제공 keywords로 결과가 없어) 질문 표면 추출로 대체 검색한 것이다 — 이 경우 핵심 절차·근거 조문이 누락됐을 수 있으므로, 1단계 키워드 추론을 보강하여 keywords와 함께 suggest_review_sources를 다시 호출한 뒤 그 결과(keyword_source=='client')로 검토를 진행할 것. degraded 응답의 candidates만으로 결론을 내지 말 것. 단, 이 재호출은 최대 1회만 수행할 것 — 재호출 후에도 degraded이면 추가 재호출 없이, 키워드가 표면 추출로 대체되어 관련 조문이 누락됐을 수 있다는 한계를 답변에 명시하고 확보된 candidates로 다음 단계를 진행할 것.
   - recommended_review_order는 기본 검토 순서로 삼되, 후보가 적으면 3단계에서 보완할 것.
   - returned·truncated·note·overflow_truncated도 확인할 것: truncated가 true이면 candidates에서 밀린 조문이 overflow_candidates에 제목(label)·provision_id로 나열되니, 관련 있어 보이는 항목은 candidates와 중복 제거 후 4단계에서 그 provision_id로 get_provision_detail을 직접 호출해 확인할 것. overflow_truncated가 true이거나 쟁점상 후보가 부족하면 recommended_review_order의 전체 문서 목록을 기준으로 3단계에서 search_provision으로 추가 보완할 것.

3. search_provision(query=...)으로 추가 검색 및 주제별 cross-check
   - 핵심 키워드, 법령상 유사어, 절차어(승인, 통보, 보고, 협약변경, 정산, 제재 등)로 검색할 것.
   - suggest_review_sources 후보와 중복 제거 후 통합할 것.
   - 주제별 Tier 2 cross-check (해당 시):
     연구개발비/예산/비목/집행 → rnd_funding_standard | 동시수행/과제 수 → simultaneous_research_limit
     시설/장비/기자재 → facility_equipment_standard | 연구노트/실험노트 → research_note_guideline
   - 공통/사업 행정규칙 cross-check (해당 시):
     정보등록/IRIS/NTIS → rnd_info_processing | 보안과제/연구보안 → rnd_security_measures | 성과평가/성과관리 → performance_eval_act + decree
     기술료/정부납부기술료 → tech_fee_integrated · sme_tech_fee(국방 R&D 기술료는 defense_tech_fee_notice) | 정보통신·방송 R&D → ict_rnd_management · ict_research_ethics
     보건의료기술 R&D/병원연구/의료기기 → health_tech_act · health_tech_decree · health_tech_rule · health_rnd_operating
     감염병·질병관리·질병관리청 R&D → kdca_rnd_management · kdca_agency_designation · kdca_facility_equipment · kdca_relay_operating
     산학협력/산학연협력/기술지주회사/협력연구소 → sanhak_act · sanhak_decree · sanhak_rule | 연구윤리/연구부정행위/연구진실성 → research_ethics_guideline
     기업부설연구소/연구개발전담부서/연구소 인정 → corp_lab_act · corp_lab_decree · corp_lab_rule
     연구산업/연구개발서비스업/연구장비산업 → research_industry_act · research_industry_decree · research_industry_rule
     연구실 안전/안전점검/정밀안전진단/연구실 사고 → lab_safety_act · lab_safety_decree · lab_safety_rule
     국방 R&D/국방연구개발/국방과학기술/국방기술/방위사업청/방사청/무기체계 연구개발/미래도전국방기술 → defense_tech_act · defense_tech_decree · defense_tech_rule · defense_rnd_guideline · defense_tech_fee_notice · defense_future_challenge_guideline · defense_standard_agreement(협약 조문) · defense_facility_equipment(국방 연구시설·장비)
     자율주행 R&D/자율주행기술개발혁신사업/자율주행 사업단 → kt_autonomous_driving(국토교통부 자율주행 사업단·과제 운영) · kt_rnd_operations(국토교통부 소관 공통 운영규정)

4. 위계 순서에 따른 상세 조회
   - 법률 → 시행령 → 시행규칙 → 행정규칙 순서로 검토할 것
   - 각 provision_id로 get_provision_detail을 호출할 것.
   - content는 OpenAPI 원문을 그대로 사용할 것.
     - OpenAPI로부터 입수한 조문의 원문을 임의로 수정(요약, paraphrase 등)하지 말 것.
     - OpenAPI로부터 입수한 조문의 항·호·목 번호를 유지할 것.
     - 단, content_format이 plain_text_verbatim이 아닌 경우(예: oversized_pointer, external_file_only)에는 그 content가 규정 원문이 아니라 안내 텍스트이므로 근거로 인용하지 말고, attached_file_url·document_source_url의 공식 원문을 확인할 것.
   - 규정의 조문·별표 본문은 임의 웹검색 결과나 law.go.kr 직접 열람 등 외부 웹에서 가져와 대체·보충하지 말고 get_provision_detail이 반환한 content로 확인할 것. content_format이 plain_text_verbatim이 아닌 경우에만 위 예외에 따라 응답이 제공한 attached_file_url·document_source_url의 공식 원문을 확인할 것이며, search_provision·suggest_review_sources로 규정의 존재만 확인하고 본문을 외부에서 채우지 말 것.
   - 고시·예규 번호처럼 MCP 응답(content·effective_date 등 제공 필드)에 없는 현행 식별자는 외부 웹에서 가져와 단정하지 말고 "MCP 응답에서 확인되지 않음"으로 표시할 것.
   - 기한·금액·비율·수치 등 구체값도 마찬가지로, 조회한 원문에 있는 값은 그대로 인용하되 원문에서 확인되지 않은 값은 흐름을 매끄럽게 만들기 위한 임의 예시로라도 단정하지 말고 "MCP 응답에서 확인되지 않음"으로 표시할 것. 감면율·요율·기한처럼 조건에 따라 값이 나뉘는 구체값을 표·목록이나 한 문장으로 압축할 때는 각 조건과 값의 대응을 원문과 같게 유지하고, 괄호·단서 등 한정어가 원문에서 어느 조건 또는 값에 귀속되는지 확인해 배치하며, 대응이 불확실하면 원문 구조대로 나눠 표시할 것.
   - 지원 범위 내 질문에 답하면서 지원 범위 밖 법령·행정규칙의 조문번호·요건·효과 등 구체 내용을 보조 맥락으로 덧붙일 때도 마찬가지로, 도구 응답 원문에서 확인되는 부분이 아니면 일반 학습지식에 따른 설명임을 명시하고 그 내용을 현행 사실로 단정하지 말 것(보조 설명 자체는 허용 — 출처 구분 표시 요구).
   - 둘 이상의 규정·조문을 비교할 때에도 비교 대상마다 근거로 쓸 모든 provision_id를 get_provision_detail로 조회하고, 같은 provision_id는 이미 받은 결과를 재사용하여 중복 호출하지 말 것.
   - (선택) 실무 해설·세부 절차·Q&A가 유용한 경우 search_manual·get_manual_section으로 「국가연구개발혁신법 매뉴얼」(본권·별권 3 제재처분 가이드라인·별권 2 기술료 제도 매뉴얼·별권 1 학생인건비통합관리 제도 매뉴얼·별권 4 연구시설・장비비 통합관리제 운영・관리 매뉴얼)·「국가연구개발 과제평가 표준지침」·「국가 R&D 연구비 부적정집행 사례집」(KAIA) 해설·사례를 참조할 것. 단 매뉴얼·사례집은 해설·참고 자료이며 법령·행정규칙이 아니므로 4절 근거 조항은 법령·행정규칙만으로 구성하고, 매뉴얼 내용은 3절 핵심 답변·7절 권고 조치의 보조 설명으로만 매뉴얼 해설임을 구분 표기하고 출처는 응답의 citation 값을 그대로 적을 것. 매뉴얼과 법령·행정규칙 내용이 다르면 법령·행정규칙 원문 우선. 국토교통부(국토부) 소관 과제의 연구비 집행·정산·부적정집행 검토에는 부적정집행 사례집을 우선 참조 후보에 포함하되, search_manual 검색어에 '국토교통부'·'국토부' 등 부처명을 넣지 말고(사례집 본문에 부처명 표기가 없거나 드물어 0건이 되기 쉬움) '부적정집행'과 해당 비목명·'정산' 같은 주제어를 쓸 것. 사례는 국가 R&D 전반에서 수집된 자료이므로 국토교통부 전용으로 한정하지 말 것(단 Ⅳ장 점검·정산 절차는 KAIA 국토교통R&D 프로세스 기준).

5. 참조 조항 추적
   - 조문이 "제X조에 따라", "시행령 제X조", "별표", "고시로 정하는" 등을 참조하면 해당 조항도 조회할 것.
   - 별표(BP)는 행정규칙·시행령 모두 get_provision_detail로 조회 가능하다(v0.2). 소형 별표는 본문 전문이 오지만, 대용량 별표는 content_format이 oversized_pointer/external_file_only로 본문이 미수록될 수 있으니 위 4단계의 content_format 규칙(plain_text_verbatim이 아니면 인용 금지)을 따를 것.
   - 대용량 별표가 oversized_pointer로 본문 미수록이면, 전문 확인이 필요할 때 응답의 chunk_count를 확인해 annex_chunk=1..chunk_count로 재호출하여 별표 본문을 줄 경계 분할 청크(원문 그대로)로 확인할 것(v0.20.0·별표 BP 전용·기본 미지정). 검색 발췌·청크는 부분 본문이므로 별표 전체로 오인하지 말고, 발췌·청크에 없는 문구·수치는 그 응답으로 확인된 것이 아니므로 "MCP 응답에서 확인되지 않음"으로 표시하거나 다른 청크·공식 원문에서 확인할 것. 청크 경계는 개정 시 달라질 수 있음(effective_date 확인). 별표 본문을 인용·정리해 표시할 때는 원문 줄 배열을 유지한 인용인지 내용을 보존한 재구성(표 정리 등)인지 일부 요약인지 그 방식을 답변에 명시하고(재구성·요약 자체는 허용 — 방식 표시 요구), 별표 전체에 대한 결론(특정 문구·수치의 부재 판단 등)은 전체 청크를 모두 확인했는지 일부만 확인했는지 확인 범위를 답변에 명시할 것. 이 방식 표시는 별표 내용만 정리한 표·목록뿐 아니라 별표 내용을 조문 등 다른 본문과 섞어 하나의 표·목록으로 정리한 경우에도 적용되며, 무엇을 기준으로 정리했다는 설명만으로는 부족하니 별표 사용 부분이 인용·재구성·요약 중 어느 방식인지 명시할 것. 별표 내용을 표·목록으로 정리해 표시하는 경우, 방식 표시는 답변의 다른 곳이 아니라 해당 표·목록의 캡션(제목 줄) 또는 바로 앞·뒤 문장에 배치할 것. 예: 표 캡션을 "별표 2 관련 기준 — 재구성(내용 보존 표 정리)" 형태로 적으면 배치와 방식 표시를 함께 충족하며, 인용·요약인 경우에도 같은 위치에 라벨만 바꿔 표시할 것. 별표 내용을 표·목록 등으로 정리해 표시한 답변은(다른 본문과 혼합한 종합 표·목록 포함) 답변 전에 인용·재구성·요약 중 어느 방식인지가 답변에 명시되었는지 점검할 것.
   - 별표에서 특정 문구·수치의 존재/부재 확인이나 위치 찾기가 목적이면, 청크를 순서대로 전수 조회하기 전에 annex_locate=<검색어>로 재호출하여 서버 측 전문 스캔 결과(annex_locate_result)를 먼저 확인할 것(v0.21.0·대용량 별표 BP 전용·기본 미지정). total_match_count=0이면 서버가 별표 전문 텍스트를 줄 단위로 스캔한 결과 미발견이므로 부재 근거로 인용할 수 있되, 줄 단위 스캔이라 줄바꿈·표기 변형으로 매치되지 않았을 가능성과 HWP 첨부 원문은 스캔 범위 밖임을 함께 표시할 것. 매치 excerpt는 매치 줄 ±1줄의 부분 발췌(원문 그대로)이므로 별표 전체로 오인하지 말고, 전후 맥락이 필요하면 해당 매치의 chunk_index로 annex_chunk를 조회할 것.
   - 별표 상세 응답에 dependent_article_hints가 있으면, 힌트에 적힌 조문을 같은 문서에서 get_provision_detail로 함께 조회할 것. 힌트는 별표 제목에서 뽑은 미검증 단서이므로 힌트 자체를 근거로 인용하지 말고, 조회된 조문 원문만 근거로 삼을 것. 이 동반 조회는 힌트에 적힌 조문 1단계까지만 자동 수행하고, 그 조문에서 이어지는 참조는 본 5단계의 일반 규칙에 따를 것.
   - 별표 번호나 가지번호가 불확실하면 BP provision_id를 추측해 호출하지 말 것. 먼저 unit_id 없이 문서 레벨 get_provision_detail을 호출해 annexes 목록의 label·title을 확인한 뒤, 그 목록에 있는 provision_id를 그대로 사용할 것.
   - 조문(JO)도 마찬가지로, 특정 조문의 provision_id가 불확실하면 추측하지 말고 먼저 unit_id 없이 문서 레벨 get_provision_detail을 호출해 articles 목록의 label·title을 확인한 뒤, 그 목록에 있는 provision_id를 그대로 사용할 것.
   - '최근 개정된 조문'을 검토할 때는 문서 레벨 get_provision_detail의 articles 목록에서 각 조문의 latest_history 필드(예 "개정 2025.12.30(공포)")로 최근 변경 조문을 찾아 그 조문을 조회할 것. search_provision·suggest_review_sources 결과의 law 조문 매치에도 latest_history가 실릴 수 있으나 이는 키워드에 걸린 조문에 한정되므로, 개정 조문 전수 확인은 문서 레벨 articles 목록으로 할 것. 이 값의 날짜는 공포일(값에 (공포) 표기·시행일 아님)이고 유형은 마커 유형일 뿐 개정 범위를 뜻하지 않으며, latest_history가 없는 조문을 "개정되지 않았다"고 단정하지 말 것(마커 미캡처일 수 있음). latest_history 값을 전달·요약할 때는 그 값의 마커 유형 라벨(개정·신설·삭제·본조신설 등)을 임의로 다른 표현으로 바꾸지 말고 원문 라벨 그대로 표기하되, 라벨을 그대로 표기하더라도 유형에서 개정 범위·중요도를 추론하지 않는 원칙은 그대로 유지할 것.
   - '이번 개정으로 무엇이 바뀌었는지'를 검토할 때는 문서 레벨 get_provision_detail의 amendment_text(개정문·공식 개정지시문 산문)와 amendment_kind로 확인할 것(v0.17.0 law·v0.19.0 admrul — 양 트랙). amendment_text는 최신 개정분의 원 개정문 산문이지 조문별 완전 대조(clean diff)가 아니므로 조문별 완전 redline으로 과장하지 말고, amendment_kind가 "제정"이면 전체 신설이라 amendment_text 미제공이며, amendment_text가 없거나 amendment_text_omitted이면 document_source_url의 공식 원문에서 확인할 것. ★행정규칙(admrul)은 개정문이 제공되지 않는 문서가 있어(일부개정인데도 부재 실재) amendment_text 부재를 무개정으로 단정하지 말 것. 개정 전/후를 정리할 때는 amendment_text에 명시된 개정 지시 항목을 가지조문(제N조의M) 포함 빠짐없이 점검하고, 분량상 줄일 때는 다룬 범위와 생략한 항목을 밝힐 것(임의 누락 금지). 개정후 대체문 인용·정리 시 그 안의 근거 법률 인용구는 법명·조문번호·'에 따른' 연결어를 포함한 원문 단위 그대로 보존할 것(예: '「법명」 제N조에 따른 기관' 패턴 등에서 '제N조에 따른' 탈락 금지). 조문번호를 탈락시키거나 기관명만으로 축약하지 말고, 여러 조문 나열 정리 시에도 각 항목에서 동일하게 유지할 것. 요약·정리 자체는 허용되나, 근거 법률 인용구를 옮긴 경우 답변 전 원문의 조문번호와 '에 따른' 연결어 누락 여부를 점검할 것. 제정·타법개정의 배경으로 도구 응답으로 확인되지 않은 전신 법령명·연혁은 단정하지 말 것(미확인 배경은 추정임을 밝히거나 생략).
   - '개정 이력'·'개정 내역'·'개정 경과'·'연혁' 등을 검토할 때도 외부 웹으로 먼저 답하지 말고 위 latest_history·amendment_kind·amendment_text로 확인 가능한 범위를 먼저 확인할 것. 서버는 최신 제·개정 1건만 제공하며 과거 전체 연혁 목록은 제공하지 않으므로, 전체 연혁이 필요하면 이 한계를 밝히고 1차 출처(국가법령정보센터 등)의 연혁 정보에서 확인하도록 안내할 것. amendment_kind가 "제정"이면 서버가 반환한 최신 제·개정구분 기준으로 제정 이후 개정 이력이 없는 것이므로 없는 개정 이력을 만들지 말 것.
   - 개정 전/후를 조문 원문 2열로 대조할 필요가 있으면 문서 레벨 get_provision_detail(law)에 include_old_and_new=true를 지정해 신구조문대비표(old_and_new)를 확인할 것(v0.18.0·law 한정·기본 미조회). 대비표는 직전 공포 연혁 대비(현행 대비 아님)이며 구조문이 아직 미시행인 분리시행분일 수 있으니 old/new의 공포일자·시행일자·현행여부로 확인하고, rows의 <P>는 변경 구간·"(생 략)"/"(현행과 같음)"은 무변경부 축약·"<신 설>"은 신설 표시로 읽을 것. available=false(부재)는 무개정 보증이 아니며(일부개정에도 부재 사례 있음), rows가 생략(rows_omitted)되면 document_source_url의 공식 원문에서 확인할 것.
   - 참조 조항 확인 없이 결론을 확정하지 말 것.

6. 조문 요건 해석, 사실관계 분석, 상위 규정 우선 원칙
   - 조문 요건 해석
     - 재량·의무 구분: "할 수 있다"는 재량, "하여야 한다"는 의무로 판단할 것.
     - 선택·병렬 구분: "하거나"와 "하고"를 혼동하지 말 것.
     - 조회한 조문에서 의무·재량·금지·예외·선택·병렬 요건을 분리하여 정리할 것.
   - 사실관계 분석
     - 정리한 조문 요건과 사용자가 제시한 사실관계를 1:1로 대응시킬 것.
     - 대응 결과를 다음으로 구분할 것: 충족 확인 / 불충족 확인 / 사실 부족 / 규정 미확인 / MCP 범위 밖.
     - 규정상 근거가 불명확한 경우, 가능성·한계·추가 확인 필요를 분리하여 작성할 것.
   - 상위 규정 우선 원칙
     - 규정 간 충돌 시 상위 규정 우선 적용
     - 일반법·특별법 관계는 사안 특성에 따라 판단.

== 최종 출력 형식 ==
- 아래 1~7절의 제목과 순서는 항상 그대로 사용할 것. 8절(절차 흐름)은 조건부 절이므로, 아래 8절의 조건에 해당할 때만 7절 뒤에 추가하고, 해당하지 않으면 제목도 작성하지 말 것.
- 중요한 정보 위주로 답변을 구성할 것.
- 불필요한 정보가 답변에 포함되지 않도록 주의할 것.
  - 단, 근거 조항의 원문 인용은 생략·요약하지 말 것.

## 【규정 검토 결과】

### 1. 상황 요약
[1-2문장으로 핵심 사실과 쟁점을 요약할 것.]

### 2. 검토 규정
- Tier 1 법률·시행령·시행규칙: [규정명 목록]
- Tier 2 행정규칙: [규정명 목록, 없으면 "해당 없음"]

### 3. 핵심 답변
- 결론: [허용/불가/승인 필요/보고 필요/추가 확인 필요 등으로 명확히 기재]
- 이유: [1-3문장으로 근거 조항과 연결]

### 4. 근거 조항
각 근거는 아래 형식을 반복할 것.
- [규정명] [조문번호] — provision_id: [provision_id]
  - 원문:
    > [get_provision_detail의 content를 verbatim 인용]
  - 적용: [이 조항이 어느 판단단위(행위·주체·절차·금액·기간)에 적용되며, 사실관계를 충족/불충족하는지]
  - 표현 판단: [의무/재량/금지/예외/선택·병렬 중 표시]

### 5. 위계 및 충돌 검토
- 상위법 우선: [상위법과 하위 규정 관계]
- 충돌 여부: [충돌 없음/충돌 가능/추가 확인 필요]

### 6. 쟁점·결손 분석
- 조문상 불명확한 부분: [없으면 "해당 없음"]
- 사용자가 제공하지 않은 필수 사실: [없으면 "해당 없음"]
- MCP 미커버 자료 확인 필요: [없으면 "해당 없음"]
- 위 각 항목이 결론에 미치는 영향: [예: "사실 부족으로 단정 불가" 등]
- 가지조문(제N조의M, 예: 제7조의2)은 v0.14.0부터 검색·상세조회 지원 — 누락 아님(문서레벨 get_provision_detail의 articles 목록에서 provision_id 확인 가능)

### 7. 권고 조치
- 규정상 확인된 후속 절차·승인·보고·문서화 조치만 기재할 것.
- 법률 판단이 필요한 사안(징계·소송·제재 비례성·승소 가능성)은 변호사 자문 권고를 표시할 것.

### 8. 절차 흐름
[검토 결과의 핵심이 둘 이상의 시간순 단계(예: 신청 → 협의 → 승인 → 보고) 또는 [예]/[아니오] 조건 분기를 포함하는 경우에만 작성할 것. 단순 정의·단일 조항 설명·단일 승인/보고 필요 여부 판단이면 본 절 전체(제목 포함)를 생략할 것.]
- 흐름은 언어 지정 없는 Markdown 코드블록 안에 번호 단계와 화살표(→)로 작성하여 모든 클라이언트에서 읽히도록 할 것.
- 각 단계에는 근거 규정명·조문번호를 함께 표시하고, 4절 근거 조항 및 7절 권고 조치와 일치시킬 것.
- 4절 또는 7절에서 직접 확인되지 않은 접수·검토·결재·통보 등 일반 실무 단계는 흐름을 매끄럽게 만들기 위해 임의로 추가하지 말 것.
- 조건 분기는 [예]/[아니오]로 표시하고, 규정상 선후관계가 확인되지 않으면 순서로 단정하지 말고 "추가 확인 필요"로 표시할 것.

### 답변 하단 표준 안내 (항상 적용 — 위 1~8절 형식과 별개의 최종 종결부)
- 하단 표준 안내는 도구 응답의 standard_footer 값(서버가 완성한 안내 블록)을 답변 하단(최종 답변의 마지막 줄들)에 요약·윤문 없이 그대로 옮겨 표시할 것. 직접 조립하지 말 것.
- 값 선택은 마지막 응답 값을 고르지 말고 최종 답변에 매뉴얼 내용을 인용했는지로 정할 것: 인용했으면 매뉴얼 응답 manual_meta의 값(처음 두 줄에 법령·매뉴얼 원문 확인 안내가 이미 포함 — 규정 응답의 값을 덧붙이지 말 것), 인용하지 않았으면 규정 도구(search_provision·suggest_review_sources·get_provision_detail) 응답의 값(여러 응답에 같은 값이 있으면 아무 하나만).
- 선택에 맞는 standard_footer가 없는 경우(구버전 응답·크기 상한으로 생략된 응답)에만 "※ 정확한 최종 확인은 국가법령정보센터(law.go.kr)의 관련 규정 원문을 기준으로 해주시기 바랍니다."와 "※ 「국가연구개발혁신법 매뉴얼」 등 연구행정 관련 매뉴얼 원문은 KISTEP 홈페이지(www.kistep.re.kr)에서 확인하시기 바랍니다."를 이 순서로 직접 표시하고, 매뉴얼 내용을 인용했다면 이어서 "※ 매뉴얼 해설 부분은 「국가연구개발혁신법 매뉴얼」을 참고한 설명입니다. 매뉴얼은 법령·행정규칙이 아니며, 내용이 다를 때는 법령·행정규칙 원문이 우선합니다."와 매뉴얼 응답 manual_meta의 notice 값 한 줄을 이 순서 그대로 추가할 것.
- footer 블록은 답변당 정확히 1개만 표시하고, 여러 값을 연결·반복하거나 같은 취지의 면책·확인 문구를 별도로 만들어 중복 부착하지 말 것.

Note: The {situation} above is a placeholder that is automatically replaced with the user's review context/question during actual execution. It is automatically entered when run with the MCP prompt (review_regulation).


Related MCP server: LexGuard MCP

Usage Examples

Below are actual usage scenarios (2x speed GIF).

Question:

  • What is the maximum amount of research allowance that can be budgeted? I'm also curious whether one person can receive the entire amount.

    • (Follow-up) What do I need to do to increase the research allowance?

Inquiry on identifying research allowance-related regulations and contract change procedures (2x speed)

Example 2) Comparison of similar regulations

Question:

  • Please explain how the 'Korea Disease Control and Prevention Agency National R&D Facility and Equipment Management Regulations' differ from the cross-ministry common 'Standard Guidelines on the Management of National R&D Facilities and Equipment'.

Comparison of KDCA facility/equipment regulations and cross-ministry facility/equipment regulations (2x speed)

Installation

Prerequisite: Get an API Key from the National Law Information Center

This MCP server uses the National Law Information Center OpenAPI to retrieve research administration-related regulations.

If applying for the first time (free / takes about 5 minutes)

  1. Go to https://open.law.go.kr → Sign up

  2. Log in → [My Page] → [API Application] → Apply for the "Law" category → Wait for 'approval'

  3. API Key issuance (How to check the key: Law Information Center → My Page → API Authentication Key Management)

If you already have a key → Proceed to the next step


Installation

Option 1: Claude.ai (Using Claude on the web)

  1. Log in to claude.ai

  2. Click 'Customize' in the left sidebar

  3. Click 'Connectors'

  4. Click the '+' next to the magnifying glass

  5. Click 'Add custom connector'

  6. Name: korean-rnd-regs-mcp

  7. url: Paste the address below (replace mcp?oc=XX → XX with the API key received from the National Law Information Center)

https://mcp.rndmanagers.org/mcp?oc=이_부분을_본인_국가법령정보센터_API키로_교체
  1. Click 'Add'

  2. Click the created 'korean-rnd-regs-mcp'

  3. On the right, change the value of 'tool permissions' from 'Needs approval' to 'Always allow'

  4. Done

Installation Check

When you enter the prompt below in a new chat, if 'status=ok' is displayed, installation is complete

korean-rnd-regs-mcp 서버의 health 테스트를 진행해줘.
Update → Automatic update (always uses the latest version)

Option 2: Claude Code (Using Claude in the terminal)

  1. uv must be installed.

  • Check whether uv is installed in the terminal → When you enter the command below, the uv version should be displayed.

  uv --version
  • If uv is not installed → Enter the command below to install uv

    • Windows

    powershell -c "irm https://astral.sh/uv/install.ps1 | iex"
    • Mac

    brew install uv
  1. Start Claude Code → Enter the command below

/plugin marketplace add smilemin07/korean-rnd-regs-mcp
  1. Enter the command below

/plugin install korean-rnd-regs@korean-rnd-regs-marketplace
  1. In the additional selection screen, choose according to the purpose of using mcp

  • If you want to use it only in the current working directory → Local

  • If you want to use it across the entire system → Global

  1. Enter the command below

/reload-plugins
  1. Done

Installation Check

When you enter the prompt below in Claude Code, if 'status=ok' is displayed, installation is complete

korean-rnd-regs-mcp 서버의 health 테스트를 진행해줘.
Update
  1. How to check if a new version exists → Enter the command below in Claude Code

/plugin marketplace update korean-rnd-regs-marketplace
  1. How to update → Enter the commands below in Claude Code in order (enter one line at a time)

/plugin update korean-rnd-regs@korean-rnd-regs-marketplace
/reload-plugins
  1. Update check → Use the health test in 'Installation Check' above to verify that the version value is the latest. If the version remains the same (uvx cache), run the command below in the terminal and then restart Claude Code

uvx --refresh korean-rnd-regs-mcp --version

Option 3: Codex (Using ChatGPT in the terminal)

  1. In the Terminal (before using Codex), enter the command below

codex mcp add korean-rnd-regs --url "https://mcp.rndmanagers.org/mcp?oc=이_부분을_본인_국가법령정보센터_API키로_교체"
  1. Start Codex

Installation Check

When you enter the prompt below in Codex, if 'status=ok' is displayed, installation is complete

korean-rnd-regs-mcp 서버의 health 테스트를 진행해줘.
Update → Automatic update (always uses the latest version)

Option 4: ChatGPT (Using ChatGPT on the web — Developer mode)

If you enable developer mode in the web version of a paid ChatGPT plan (Plus, Pro, etc.), you can connect this server (verified with a real account in 2026-08 — confirmed that provision/manual lookup and the notice at the bottom of answers all work properly).

⚠️ Caution: In developer mode, you can connect custom MCP apps that have not been verified by OpenAI. Before connecting any server, check the operating entity and server address, and only connect to servers you trust. Untrusted servers can increase security risks such as prompt injection.

  1. Log in to chatgpt.com → Profile icon → Settings → Security and Login → Turn on Developer mode

  2. Sidebar Plugins(Create) → Enter the following and save

    • Name: Any name you want (e.g., korean-rnd-regs-mcp)

    • MCP server URL: https://mcp.rndmanagers.org/mcp?oc=Replace_this_part_with_your_National_Law_Information_Center_API_key (do not add a / at the end)

    • Authentication: No authentication

  3. If 7 tools are displayed, the connection was successful

  4. Use in conversations: Type the registered app name in the menu of the input field to select it (it does not appear in the default list)

Update → Automatic update (reflected when the server is redeployed. Only updates that add new tools require refreshing the tools in the app settings)

Other (Environments with restricted web access)

  • Gemini (web): The general chat does not have a feature to register custom MCP URLs, so it cannot be used (as of 2026-07). However, Gemini-series CLI and API support MCP.


Supported Regulations (66 total)

Tier 1 — Core laws, enforcement decrees, and enforcement rules [cross-ministry (MSIT, etc.) R&D application] (3):

ID

Name

Type·Effective Date

innovation_act

National R&D Innovation Act

Law (2026-08-20)

innovation_decree

Enforcement Decree of the National R&D Innovation Act

Presidential Decree (2026-08-20)

innovation_rule

Enforcement Rule of the National R&D Innovation Act

MSIT Ordinance (2026-08-20)

Sector — MSIT R&D-related regulations (12):

ID

Name

Type·Effective Date

msit_rnd_processing

Regulations on the Processing of Science and Technology R&D Projects under the Ministry of Science and ICT

Administrative Rule (2023-08-24)

ict_rnd_management

Information and Communication·Broadcasting R&D Management Regulations

Administrative Rule (2025-05-13)

ict_research_ethics

Regulations on Securing Research Ethics Integrity in Information and Communication·Broadcasting

Administrative Rule (2024-10-31)

corp_lab_act

Act on Support for R&D Activities of Corporate Research Institutes, etc.

Law (2026-02-01)

corp_lab_decree

Enforcement Decree of the Act on Support for R&D Activities of Corporate Research Institutes, etc.

Presidential Decree (2026-02-01)

corp_lab_rule

Enforcement Rule of the Act on Support for R&D Activities of Corporate Research Institutes, etc.

MSIT Ordinance (2026-02-01)

research_industry_act

Research Industry Promotion Act

Law (2021-10-21)

research_industry_decree

Enforcement Decree of the Research Industry Promotion Act

Presidential Decree (2024-06-01)

research_industry_rule

Enforcement Rule of the Research Industry Promotion Act

MSIT Ordinance (2024-06-01)

lab_safety_act

Act on the Establishment and Promotion of a Safe Research Environment

Law (2026-05-20)

lab_safety_decree

Enforcement Decree of the Act on the Establishment and Promotion of a Safe Research Environment

Presidential Decree (2026-05-20)

lab_safety_rule

Enforcement Rule of the Act on the Establishment and Promotion of a Safe Research Environment

MSIT Ordinance (2026-02-01)

Sector — Defense Acquisition Program Administration (DAPA) defense R&D-related regulations (8):

ID

Name

Type·Effective Date

defense_tech_act

Defense Science and Technology Innovation Promotion Act

Law (2024-07-10)

defense_tech_decree

Enforcement Decree of the Defense Science and Technology Innovation Promotion Act

Presidential Decree (2026-07-01)

defense_tech_rule

Enforcement Rule of the Defense Science and Technology Innovation Promotion Act

Ministry of National Defense Ordinance (2021-04-01)

defense_rnd_guideline

Defense Technology R&D Work Processing Guidelines

Administrative Rule (2026-02-10)

defense_tech_fee_notice

Notice on the Methods of Calculating, Collecting, and Procedures for Defense Science Technology Fees

Administrative Rule (2026-02-10)

defense_future_challenge_guideline

Future Challenge Defense Technology R&D Work Processing Guidelines

Administrative Rule (2026-02-10)

defense_facility_equipment

Regulations on the Management of Defense R&D Facilities and Equipment

Administrative Rule (2026-02-10)

defense_standard_agreement

Standard Agreement for Weapon System R&D

Administrative Rule (2026-01-02)

Sector — MOLIT R&D-related regulations (5):

ID

Name

Type·Effective Date

sector_kt_act

Construction and Transportation Science and Technology Promotion Act

Law (2026-02-01)

sector_kt_decree

Enforcement Decree of the Construction and Transportation Science and Technology Promotion Act

Presidential Decree (2026-08-20)

sector_kt_rule

Enforcement Rule of the Construction and Transportation Science and Technology Promotion Act

MOLIT Ordinance (2018-06-08)

kt_rnd_operations

Regulations on the Operation of R&D Projects under the Ministry of Land, Infrastructure and Transport

Administrative Rule (2026-07-08)

kt_autonomous_driving

(MOLIT) Autonomous Driving Technology Development Innovation Project Operation and Management Regulations

Administrative Rule (2026-07-08)

Sector — MOTIE R&D-related regulations (7):

ID

Name

Type·Effective Date

industry_tech_act

Industrial Technology Innovation Promotion Act

Law (2026-06-03)

industry_tech_decree

Enforcement Decree of the Industrial Technology Innovation Promotion Act

Presidential Decree (2026-06-03)

industry_tech_rule

Enforcement Rule of the Industrial Technology Innovation Promotion Act

MOTIE Ordinance (2026-06-03)

industry_tech_operating

Common Operating Guidelines for Industrial Technology Innovation Projects

Administrative Rule (2024-12-30)

industry_tech_security

Security Management Guidelines for Industrial Technology Innovation Projects

MOTIE Notice (2018-04-30)

industry_tech_evaluation

Technology Development Evaluation Management Guidelines for Industrial Technology Innovation Projects

MOTIE Directive (2024-12-30)

tech_fee_integrated

Integrated Guidelines on the Collection and Management of Technology Fees

MOTIE Notice (2025-04-07)

Sector — MSS R&D-related regulations (5):

ID

Name

Type·Effective Date

sme_tech_act

Small and Medium Enterprise Technology Innovation Promotion Act

Law (2026-07-01)

sme_tech_decree

Enforcement Decree of the Small and Medium Enterprise Technology Innovation Promotion Act

Presidential Decree (2026-07-01)

sme_tech_rule

Enforcement Rule of the Small and Medium Enterprise Technology Innovation Promotion Act

MSS Ordinance (2020-07-30)

sme_rnd_operating

Operating Guidelines for SME Technology Development Support Projects

Administrative Rule (2026-01-21)

sme_tech_fee

Technology Fee Management Regulations for SME Technology Development Support Projects

MSS Notice (2026-04-02)

Sector — MOHW R&D-related regulations (4):

ID

Name

Type·Effective Date

health_tech_act

Health and Medical Technology Promotion Act

Law (2026-02-12)

health_tech_decree

Enforcement Decree of the Health and Medical Technology Promotion Act

Presidential Decree (2026-02-01)

health_tech_rule

Enforcement Rule of the Health and Medical Technology Promotion Act

MOHW Ordinance (2024-07-17)

health_rnd_operating

Operation and Management Regulations for Health and Medical Technology R&D Projects

MOHW Notice (2023-12-26)

Sector — KDCA R&D-related regulations (4):

ID

Name

Type·Effective Date

kdca_rnd_management

Korea Disease Control and Prevention Agency R&D Management Regulations

Administrative Rule (2026-05-18)

kdca_agency_designation

Notice on the Designation of Specialized Agencies for KDCA R&D Projects

Administrative Rule (2026-04-15)

kdca_facility_equipment

KDCA National R&D Facility and Equipment Management Regulations

Administrative Rule (2022-08-31)

kdca_relay_operating

(KDCA) Common Operating Guidelines for the Cross-Ministry Relay Project of National R&D Outcomes

Administrative Rule (2021-02-02)

Sector — MOE R&D-related regulations (7):

ID

Name

Type·Effective Date

sanhak_act

Act on the Promotion of Industrial Education and Industry-Academia-Research Cooperation

Law (2025-06-21)

hakjin_act

Academic Promotion Act

Law (2021-06-23)

hakjin_decree

Enforcement Decree of the Academic Promotion Act

Presidential Decree (2022-11-08)

sanhak_decree

Enforcement Decree of the Act on the Promotion of Industrial Education and Industry-Academia-Research Cooperation

Presidential Decree (2026-03-24)

hakjin_rule

Enforcement Rule of the Academic Promotion Act

MOE Ordinance (2020-10-13)

sanhak_rule

Enforcement Rule of the Act on the Promotion of Industrial Education and Industry-Academia-Research Cooperation

MOE Ordinance (2026-03-27)

research_ethics_guideline

Guidelines for Securing Research Ethics

Administrative Rule (2023-07-17)

Tier 1 — Performance evaluation-related regulations (2):

ID

Name

Type·Effective Date

performance_eval_act

Act on the Performance Evaluation and Management of National R&D Projects, etc.

Law (2023-10-31)

performance_eval_decree

Enforcement Decree of the Act on the Performance Evaluation and Management of National R&D Projects, etc.

Presidential Decree (2025-04-01)

Tier 2 — Core administrative rules (5):

ID

Name

Type·Effective Date

rnd_funding_standard

Standards for the Use of R&D Funds in National R&D Projects

Administrative Rule (2026-05-06)

simultaneous_research_limit

Standards for Limiting the Number of Concurrent National R&D Project Tasks

Administrative Rule (2021-01-01)

facility_equipment_standard

Standard Guidelines on the Management of National R&D Facilities and Equipment

Administrative Rule (2026-04-23)

research_note_guideline

Research Note Guidelines for National R&D Projects

Administrative Rule (2022-01-01)

innovation_challenge_criteria

Notice on the Designation and Classification Criteria for Innovation Challenge-type R&D Project Groups

MSIT Notice (2025-02-03)

Tier 2 — Additional common administrative rules (4):

ID

Name

Type·Effective Date

rnd_info_processing

National R&D Information Processing Standards

Administrative Rule (2026-08-20)

rnd_security_measures

National R&D Project Security Measures

Administrative Rule (2023-11-20)

large_rnd_plan_review

Regulations on the Operation of Large R&D Project Plan Reviews

MSIT Notice (2026-03-26)

build_type_rnd_screening

Guidelines for the Operation of Construction-Type R&D Project Screening

MSIT Notice (2026-05-11)

Expansion Direction: The scope of supported regulations will continue to be expanded.

  • Fields (breadth): Includes cross-ministerial common, land and transportation, industry, SME, health and medical, disease control, and education (academic promotion·industry-academia cooperation) R&D regulations → planned to continue expanding to other ministries

  • Material types (depth): Centered on statutes·administrative rules provided by the National Law Information Center OpenAPI + Innovation Act manual commentary (main volume v0.27.0·separate volume 3 Sanctions Guidelines v0.32.0·separate volume 2 Technology Royalty System Manual v0.33.0·separate volume 1 Student Labor Cost Integrated Management System Manual v0.35.0·Project Evaluation Standard Guidelines v0.38.0·KAIA Research Fund Improper Expenditure Casebook v0.43.0 added) → planned to continue expanding with additional materials

Using It Reliably

Regulation review is accurate when this MCP tool is actually invoked and retrieves the supporting provisions. If the tool is not invoked, the AI may answer based on general knowledge, and in that case, figures·requirements may differ from the actual regulations. Please refer to the following:

  • On claude.ai web, if you use multiple connectors (Google Drive·Gmail, etc.) together in one conversation, the regulation tool may not be invoked during the conversation. When reviewing regulations, temporarily turn off connectors you are not using.

  • If the answer mentions regulation content but does not cite provisions (provision_id) or provide original-text sources, the tool may not have been invoked (absence of citations is a warning signal, not a certainty) → ask again in a new conversation.

  • For important regulation reviews conducted over multiple sessions, Claude Desktop·Claude Code (stdio local installation) is more stable than web connectors — the tool is loaded per conversation (session), so the likelihood of it being dropped mid-way is lower.

  • Supported scope = 66 regulations (Innovation Act family·ministry-specific R&D·core administrative rules). Answers about regulations outside this scope are based on general knowledge, so please also check primary sources such as the National Law Information Center.

Features

7 MCP tools:

Tool

Purpose

health

Check service status·API key configuration

list_rule_sets

View the list of 66 registered regulations·hierarchy rank·document ID·governing ministry

search_provision

Keyword search in provision·appendix text → snippet + provision_id candidate list. Normal responses excluding entry errors include the complete standard footer standard_footer at the bottom of the answer (v0.40.0) + adjacent instruction standard_footer_note (display rules for list-type·0-result answers) — attached when budget (16,000 characters) allows·2-stage fallback keeping only the footer when exceeded (v0.41.0)

get_provision_detail

Verbatim lookup of a single provision/appendix text by provision_id (includes metadata to prevent LLM from inventing arbitrary subtitles). When querying by document, provides a list of appendix titles together — so the AI selects by title rather than guessing the appendix (v0.2.1). Non-error responses include the complete standard footer standard_footer at the bottom of the answer (v0.29.0)

suggest_review_sources

Natural language question → keyword extraction → candidate regulations·provisions to review + recommended review order (Act → Enforcement Decree → Enforcement Rule → Administrative Rule). When keywords are not passed together, the server's automatic extraction only trims the request clause at the end of the question ('…regulation candidates and review order', etc.), and preserves the same words when used as provision terms (v0.42.0). Normal responses excluding entry errors include standard_footer (v0.40.0) + adjacent instruction standard_footer_note (when budget allows·keeps only footer when exceeded) (v0.41.0)

search_manual

Integrated search of the 「National R&D Innovation Act Manual」 main volume + separate volume 3 「Sanctions Guidelines」 + separate volume 2 「Technology Royalty System Manual」 + separate volume 1 「Student Labor Cost Integrated Management System Manual」 + 「National R&D Project Evaluation Standard Guidelines」(25.12) + separate volume 4 「Research Facility·Equipment Cost Integrated Management System Operation·Management Manual」 + 「National R&D Research Fund Improper Expenditure Casebook」(KAIA·25.5) — clause-level matching (distinguished by source) + print-page anchor excerpts + complete citation citation (v0.39.0 series completed with separate volume 4·v0.43.0 casebook added)

get_manual_section

Lookup of manual·guideline section text (main volume 3-4 type·separate volume 3 b3-4-2 type·separate volume 2 b2-3-2 type·separate volume 1 b1-3-4 type·project evaluation standard guidelines eval-3-2 type·separate volume 4 b4-5 type [flat organization without chapters, so single level]·improper expenditure casebook case-2-2 type id) — large sections are split into chunks at page boundaries. All non-error responses include normative metadata (commentary material·statute priority)·complete citation·bottom standard notice block (v0.28.0). Sections where tables·formulas are not preserved in the extracted text include the complete notice block structure_notice (v0.34.0)

Innovation Act Manual Commentary (v0.27.0)

You can search and retrieve by section: the commentary section of the 「National R&D Innovation Act Manual」(main volume, 26.7 edition·based on the statute effective date of June 2026) (5 chapters 40 sections + 3 references·print 1332 pages), separate volume 3 「National R&D Project Sanctions Guidelines」(26.7 published set·5 chapters 22 sections + 1 appendix·print 189 pages·v0.32.0 added), separate volume 2 「National R&D Project Technology Royalty System Manual」(26.7 published set·3 chapters 14 sections + 1 appendix·print 112 pages·v0.33.0 added), separate volume 1 「Student Labor Cost Integrated Management System Manual」(26.7 published set·5 chapters 22 sections + FAQ + 3 references·print 391 pages·v0.35.0 added), separate volume 4 「Research Facility·Equipment Cost Integrated Management System Operation·Management Manual」(26.7 published set·overview + main text ⅠⅨ + attachment·print 350 pages·v0.39.0 added — with this, all volumes of the main volume·separate volumes 1~4 are included), and the 「National R&D Research Fund Improper Expenditure Casebook」(Korea Authority of Land & Infrastructure Safety (KAIA)·2025.05·105 cases by budget item + 18 FAQ items + inspection·settlement procedures·v0.43.0 added).

  • Separate volume 3 is a commentary on sanctions (participation restriction·additional sanctions surcharge·recovery) procedures·standards·re-review issues. Specific values such as participation restriction periods·additional sanctions surcharge rates are based on the original text of Enforcement Decree Appendix 6·Appendix 7 (get_provision_detail), so cross-check when citing. The original text of separate volume 3 does not specify the statute reference date, and that fact is noted in the response notice.

  • Separate volume 2 is a commentary on the technology royalty system (collection·government payment technology royalties·reduction/exemption·use) and the guidelines for calculating·verifying technology contribution at the time of agreement. Specific values such as royalty rates·payment caps·payment periods are based on the original text of Enforcement Decree Articles 38~41 (get_provision_detail), so cross-check when citing. The original text of separate volume 2 also does not specify the statute reference date, and some terms in the text (e.g., technology royalty payment obligation institution) are terms from before the Enforcement Decree amendment effective 2026-07-28.

  • Separate volume 1 is a commentary on the student labor cost integrated management system (permitted uses·accounting standards·integrated management account·payment·balance and interest handling·transfer·refund·designation and revocation of integrated management institutions) plus FAQ·designation status·standard operating guidelines·inspection materials. Specific values such as accounting standards·payment amounts·transfer deadlines are based on the original text of the 「National R&D Project Research and Development Cost Usage Standards」 (get_provision_detail·rnd_funding_standard), so cross-check when citing. The original text of separate volume 1 also does not specify the statute reference date, the notice number cited in the text (No. 2026-5) is the notation at the time of publication and may differ from the current amended version number, and the description of student labor cost interest handling is based on the structure of Article 75 of the older notice (that fact is noted in the response notice).

  • Separate volume 4 is a commentary on the research facility·equipment cost integrated management system (designation of implementing institutions·opening and operation of integrated management accounts·accounting·payment·accumulation·use·management·supervision·designation revocation). Because it has a flat organization without chapters, section ids are single-level (b4-0 overview·b4-1b4-9 main text ⅠⅨ·b4-ref-1 attachment). Specific values such as accumulation·use requirements and designation·revocation criteria are based on the original text of Chapter 7 (Articles 100~111) of the 「National R&D Project Research and Development Cost Usage Standards」 (get_provision_detail·rnd_funding_standard), so cross-check when citing. The regulation excerpts in the attachment are a snapshot of Notice No. 2023-49 (2023.12.28.) based on the original text's supplementary provisions notation, and some wording differences from the current notice have been identified (that fact is noted in the response notice).

  • The 「National R&D Research Fund Improper Expenditure Casebook」 is an educational·reference casebook published by the Korea Authority of Land & Infrastructure Safety (KAIA) (105 cases of settlement non-recognition by budget item·18 FAQ items·precautions before using research funds·regular·annual inspection/settlement procedures). Section ids are case-1-1 (Ⅰ Before You Begin)·case-2-1~case-2-10 (Ⅱ Cases by Budget Item)·case-3-1 (Ⅲ FAQ)·case-4-1 (Ⅳ Inspection·Settlement Procedures). The cases are not settlement rulings or sanction decisions on individual matters, and conclusions may vary depending on institution type·agreement·evidence; cross-check specific amounts·rates·provisions against the current original text of the 「National R&D Project Research and Development Cost Usage Standards」. The inspection·settlement procedures in Chapter Ⅳ (Ez-baro·entrusted settlement institutions·schedule) are based on KAIA land transportation R&D processes, so for projects of other ministries, confirmation from the relevant specialized institution is required (that fact is noted in the response notice).

  • Manuals·casebooks are commentary·reference materials — they are not statutes·administrative rules, and when content differs, the original text of statutes·administrative rules takes precedence. Always verify normative grounds such as obligations·deadlines·amounts using the statute tools (search_provision·get_provision_detail).

  • All manual responses include normative metadata (manual_meta) containing the edition·reference date·statute priority principle and source notation (notice).

  • Included scope: main volume commentary section + separate volume 3 「Sanctions Guidelines」 in full (print 189 pages) + separate volume 2 「Technology Royalty System Manual」 in full (print 112 pages) + separate volume 1 「Student Labor Cost Integrated Management System Manual」 main text (print 391 pages — excluding chapter title pages·blank pages) + separate volume 4 「Research Facility·Equipment Cost Integrated Management System Operation·Management Manual」 main text (print 350 pages — excluding title pages·blank pages), so all volumes of the main volume·separate volumes 1~4 are included, plus the 「National R&D Research Fund Improper Expenditure Casebook」 main text (print 6~~95 pages — excluding cover·table of contents·title pages·blank pages·v0.43.0). Appendix-related forms (published as separate PDFs from the 26.7 edition) are not included.

  • KISTEP publishes revised editions of the manual from time to time (twice in 2026: April and July) — until a revised edition is reflected, statute amendments after the manual's reference date may not be reflected in the commentary.

1 MCP prompt:

Prompt

Purpose

review_regulation

Multi-layered regulation review — applies the 66 regulations in hierarchical order to a specific case and answers with regulation review results along with citations of supporting provisions. Must include the 'korean-rnd-regs-mcp', 'regulation review' keywords in the prompt


Troubleshooting

Q0. New tools (search_manual, etc.) are not visible after a server update

After an update that adds tools to the server (new input schemas), the client may be caching the previous tool list.

  1. claude.ai web connector: Press the ⋮ menu → "Refresh tools list" in the connector detail screen to refresh only the tool list without disconnecting (recommended). If it still doesn't appear, disable → re-enable (reconnect) the connector. In either case, you must use it in a new conversation after the refresh; ongoing conversations may retain the previous tool list.

  2. Claude Desktop / Claude Code (stdio): Restart the app or run /reload-plugins, then start a new conversation.

  3. If it still doesn't appear after reconnecting, check with health whether the server version is up to date.

Q1. The AI answers "No relevant provisions found"

  1. Ask the AI to check the errors field of the tool response — API authentication failures, etc. are delivered via errors

  2. Check api_key_configured: true with the health tool

  3. Verify that LAW_API_KEY is correct on the National Law Information Center OpenAPI My Page

  4. Remove any leading/trailing whitespace from the key string

Q2. The AI invents subtitles or summaries not in the provision text

  • Add the phrase "do not edit, keep as is" or "verbatim" to the prompt

Q3. Provisions that were in effect in the past, not the currently effective ones, are output.

  • If the work proceeded without this tool (MCP) being invoked, results like the one in the question can occur.

    • Try sending the prompt with this MCP's name included.

Q4. "auth_failed" error

  • LAW_API_KEY is empty or incorrect. open.law.go.kr My Page → reissue API key

  • If using remotely (web connector), check whether the ?oc= value in the connector URL matches your own API key.

Q5. "rate_limited" error

  • Daily call limit reached. Wait 24 hours and retry. The cache (24h) reduces repeated calls for the same query

Q6. (ChatGPT) A warning appears in the tool approval window, or read-only lookups are classified as "write"

  • ChatGPT shows an approval window before tool calls, and when connecting to this server, warnings about the tool descriptions (InstructionAttempt, etc.) may be displayed together. The tool descriptions on this server include instructions to quote statute provisions verbatim without summarization or modification, and ChatGPT's automatic classifier appears to flag such output-related instructions as risk signals (the exact cause cannot be confirmed because OpenAI does not disclose the warning criteria). Please verify the server address and operating entity before connecting, then decide whether to approve.

  • All 7 tools on this server are read-only lookups, and a read-only indicator (readOnlyHint) has been assigned since v0.36.0. If you connected previously, they may be displayed as "write" until the tool list is refreshed, so refresh the tool list from the plugin (connector) management screen's "Refresh" (no re-registration or key re-entry needed).

Q7. (ChatGPT) The standard notice at the bottom of the answer is not visible

  • We observed cases in ChatGPT where the bottom standard notice (statute original text verification·manual source notice) was not displayed in answers that received search·candidate lists. Since the same question with a higher-tier model on the same connector·same server displays it normally, we judge this to be a model-specific difference where the notice fields at the end of the server's response are dropped during answer assembly (the conditions and frequency of occurrence cannot be confirmed). The server does send the notice normally, and there is no impact on the provision fields returned by the server (however, please verify the overall accuracy of the answer separately). If you need to verify the original text, please cross-check it at the National Law Information Center.


For Developers

Local Development Environment

git clone https://github.com/smilemin07/korean-rnd-regs-mcp.git
cd korean-rnd-regs-mcp
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -e ".[dev]"

Running Tests

pytest
# 643 passed (mock 기반, 네트워크 미사용)

Build

python -m build
twine check dist/*

Security

LAW_API_KEY is an authentication value for querying public statute data, issued free of charge from the National Law Information Center OpenAPI. It differs in nature from financial API keys, but please manage it so that others cannot use it without authorization.

  • If exposed, others could repeatedly make requests with the same key and affect your usage

Server Protection Scope

  • The server automatically hides API keys from tool responses, error messages, etc.

  • Communication with the statute information API uses HTTPS encryption

User Precautions

  • Do not expose your API key in chats, screenshots, SNS, etc.

  • When using remote access (HTTP mode), the API key is included in the URL — it can remain in browser history and terminal history, so do not use it on shared PCs.

  • If you host the server directly in --http mode, bind to loopback only with FASTMCP_HOST=127.0.0.1 (if unset·empty·whitespace, it binds to all interfaces 0.0.0.0 for compatibility, allowing direct port access from the same LAN without going through a tunnel·proxy). Note that this setting only reduces exposure at the kernel network interface level — it does not prevent reachability through processes on the same host that proxy connections to loopback on your behalf (e.g., Synology Tailscale's userspace networking proxies tailnet requests to 127.0.0.1), so such paths must be controlled separately with that tool's policies, such as Tailscale ACLs.

Response to Key Exposure

If your key is exposed, you can delete the existing key and issue a new one at open.law.go.kr.

Detailed security policy·vulnerability reporting: SECURITY.md


Disclaimer

This tool was developed to support regulation review work. Responsibility for judgment outcomes rests with the user.


License

Apache License 2.0

Contributing

Issues·PRs welcome: https://github.com/smilemin07/korean-rnd-regs-mcp/issues

Changelog

2026. 8. 11.: v0.48.0

  • Improved regulation search·lookup response speed (statute information API connection reuse — stabilization update, no feature·data changes)

2026. 8. 10.: v0.47.0

  • Added 2 MSIT regulations (Regulations on the Operation of Large R&D Project Plan Reviews, Guidelines for the Operation of Construction-Type R&D Project Screening / supported regulations 64 → 66)

2026. 7. 25.-8. 8.: v0.27.0-v0.46.0

  • National R&D Innovation Act Manual ('26.7) support

    • Main volume, separate volume 1 (Student Labor Cost Integrated Management System Manual), separate volume 2 (National R&D Project Technology Royalty System Manual), separate volume 3 (National R&D Project Sanctions Guidelines), separate volume 4 (Research Facility·Equipment Cost Integrated Management System Operation·Management Manual)

  • National R&D Research Fund Improper Expenditure Casebook (published by KAIA, '25.5) support

  • National R&D Project Evaluation Standard Guidelines ('25.12) support

  • Improved the issue where pre-effective regulations were provided as if they were current regulations, etc.

2026. 7. 24.-25.: v0.26.0-v0.26.1

  • Added 1 MOLIT autonomous driving R&D administrative rule ((Ministry of Land, Infrastructure and Transport) Autonomous Driving Technology Development Innovation Project Operation Management Regulations / supported regulations 63 → 64), etc.

2026. 7. 24.: v0.25.0

  • Added 3 DAPA (Defense Acquisition Program Administration) defense R&D administrative rules (Future Challenge Defense Technology R&D Work Processing Guidelines · Regulations on the Management of Defense R&D Facilities and Equipment · Standard Agreement for Weapon System R&D / supported rules 60 → 63) etc.

2026. 7. 23.: v0.24.0

  • Added 2 DAPA defense R&D administrative rules (Defense Technology R&D Work Processing Guidelines · Defense Science and Technology Fee Notice / supported rules 58 → 60)

    • Raised internal cache limit for response speed management (64→96) etc.

2026. 7. 23.: v0.23.0

  • Added support for 3 DAPA defense R&D regulations (Defense Science and Technology Innovation Promotion Act · Enforcement Decree · Enforcement Rules / supported rules 55 → 58)

    • Added a mechanism for the AI to recognize that defense R&D is governed by the Defense Science and Technology Innovation Promotion Act framework rather than the National R&D Innovation Act, etc.

2026. 7. 22.: v0.22.0

  • Added support for 3 MSIT R&D regulations (Act on the Creation of a Safe Laboratory Environment · Enforcement Decree · Enforcement Rules / supported rules 52 → 55)

2026. 7. 9.: v0.17.0

  • Improved so that recent amendments to statutes (what has changed) can be provided to the AI (newly enacted statutes are marked as "enacted" when provided), etc.

2026. 7. 5.-8.: v0.13.0-v0.16.0

  • Added support for 1 MSIT R&D regulation (Notice on the Designation and Classification Criteria for Innovative Challenge-type R&D Project Groups / supported rules 51 → 52)

  • Improved so that recent amendment history of articles (promulgation date, etc.) can be provided to the AI

  • Improved so that sub-articles (e.g., 'Article 2-2') can be output, etc.

2026. 6. 29.-7. 1.: v0.10.0-v0.12.0

  • Added support for 6 MSIT R&D regulations and 2 MOTIE R&D regulations (Act on Support for R&D by Corporate Research Institutes, etc. family / supported rules 43 → 51)

  • Improved so that provisions under 'items' below 'subparagraphs' of statutory articles are output accurately

2026. 6. 26.: v0.9.1

  • Improved search speed and stability

    • Prevention of disconnections that may occur when the number of concurrent users increases, etc.

2026. 6. 24.: v0.9.0

  • Added support for 4 MOE R&D regulations (Industry-Academia Cooperation family 3 + Guidelines for Securing Research Ethics / supported rules 39 → 43)

2026. 6. 24.: v0.8.0

  • Added support for 3 MOE R&D regulations (supported rules 36 → 39)

2026. 6. 21.-22.: v0.5.0-v0.7.0

  • Provide the issuance number and type of administrative rules (e.g., Directive No. 179) to the AI

  • Strengthened article search capability, etc.

    • Improved the issue where the AI, when looking for a specific article (e.g., Article 2 of XX Regulation), lacked information on that article's identifier and thus performed a web search and answered with the old article

      • Identifier example: Article 2 of the Common Operating Guidelines for Industrial Technology Innovation Projects → admrul:2100000251982:JO0002

2026. 6. 20.: v0.4.0-v0.4.1

  • Added support for 4 KDCA R&D regulations (supported rules 32 → 36)

  • Increased search result cache capacity (in preparation for regulation expansion), etc.

2026. 6. 20.: v0.3.0

  • Added support for 4 MOHW health and medical technology R&D regulations (supported rules 28 → 32)

  • Added a hallucination reduction mechanism for responses on unsupported regulations: when explaining regulations outside this server's scope using general knowledge, added a mechanism to guide users to verify against the primary source (National Law Information Center)

2026. 6. 17.-20.: v0.2.8-12

  • Search result relevance sorting: improved the issue where, when a question related to multiple areas was received and later results were truncated due to response size limits, the regulation most relevant to the question was cut off

  • Operational safety improvements, security hardening work, etc.

2026. 6. 14.: v0.2.7

  • Operational stability enhancement: improved the issue where, when querying multiple regulations, slow responses from some requests delayed the entire operation

2026. 6. 13.: v0.2.6

  • Reorganized supported regulations (25 → 28): removed 6 auxiliary regulations, added 9 cross-ministerial & MSIT & MOTIE & MSS R&D regulations

    • Removed auxiliary regulations (6): Anti-Corruption Act, Improper Solicitation and Graft Act, Public Interest Whistleblower Protection Act · Enforcement Decree

      • Excluded from supported regulations due to low relevance to this MCP's primary service (R&D regulation review)

    • Added cross-ministerial & MSIT & MOTIE & MSS R&D regulations (9): Performance Evaluation Act · Enforcement Decree, National R&D Information Processing Standards, National R&D Project Security Measures, MSIT Science and Technology Field Processing Regulations, ICT and Broadcasting R&D Management Regulations, ICT and Broadcasting Research Ethics Regulations, Technology Fee Integrated Guidelines (MOTIE) · SME Technology Development Technology Fee Management Regulations (MSS)

  • Added competent ministry information to the regulation list (list_rule_sets), generalized the notice that attachments other than tables (including appendices and enclosures) cannot be retrieved, etc.

2026. 6. 13.: v0.2.5

  • Expanded supported regulations (17 → 25): added 8 core MOTIE and MSS R&D regulations

    • Industrial Technology Innovation Promotion Act · Enforcement Decree · Enforcement Rules + Common Operating Guidelines for Industrial Technology Innovation Projects

    • SME Technology Innovation Promotion Act · Enforcement Decree · Enforcement Rules + SME Technology Development Support Project Operating Guidelines

2026. 6. 12.: v0.2.4

  • Reflected the amendment to the National R&D Innovation Act · Enforcement Decree · Enforcement Rules effective 2026. 6. 11. in the regulation list information

  • Corrected the issue where, when searching with multiple keywords in large tables, frequently occurring words monopolized the excerpt space and the row for the institution actually being searched for was omitted

2026. 6. 12.: v0.2.3

  • Improved so that when keyword-matched rows appear in multiple places in large tables (e.g., Table 6 indirect cost notice rate table in the R&D expense usage standards), those rows are excerpted and shown together (up to 6 places)

    • (Previous) Only the area around the first matched row was excerpted, causing the row for the institution or figure actually being searched for to be omitted

2026. 6. 11.: v0.2.2

  • Reinforced guidance text so the AI does not hit a dead end when searching for tables and forms, etc.

    • e.g., notice that forms and templates are not supported, guidance on re-search paths when table retrieval fails

2026. 6. 10.: v0.2.1

  • Added a mechanism that prompts the AI to look up tables when table references are needed to check article details

  • Improved so that sub-tables (e.g., 'Table 1-2', etc.) can be retrieved, etc.

2026. 6. 9.: v0.2.0

  • Improved so that tables in the Enforcement Decree of the National R&D Innovation Act can be loaded

2026. 6. 8.: v0.1.10

  • Added visual flow to procedural responses

    • (Previous) Even for questions where procedures or sequences were key, answers were provided only in text, making it difficult to grasp steps and branches at a glance

    • (Changed) When review results include step-by-step procedures or conditional branches, reinforced the review prompt to present a procedural flow (text diagram) with the supporting article for each step alongside the answer

2026. 6. 7.: v0.1.9

  • Strengthened article search capability

    • (Previous) Keywords for searching relevant articles were not received from the AI, so appropriate articles for regulation review could not be fetched

    • (Changed) When keywords are not received from the AI → inserted a mechanism that prompts the AI to send keywords to the MCP server

2026. 6. 7.: v0.1.8

  • Strengthened article search capability

    • (Previous) When there were many relevant articles, only the top few were returned, so the AI could not reference other relevant articles (e.g., 'prior approval procedures')

    • (Changed) Articles pushed out of the top list are also provided in 'title + identifier' form → modified so the AI can find the articles it needs, check the full text, and conduct the regulation review

2026. 6. 6.: v0.1.7

  • Strengthened article search capability

    • (Previous) Articles that coincidentally contained many common words (e.g., government-funded R&D expenses) ranked at the top, pushing out the articles actually needed for regulation review (e.g., 'amendment of agreements', 'subjects requiring prior approval')

    • (Changed) Articles whose 'titles' directly match the question keywords are searched first

2026. 6. 5.: v0.1.6

  • Strengthened article search capability

    • (Example 1) When the prompt uses the expression 'government contribution'

      • (Previous) If the word 'government contribution' did not appear in the article → the article could not be retrieved even if needed for regulation review

      • (Changed) Expand to similar-meaning words such as 'government-funded R&D expenses', 'grants', etc. for article search → reference relevant articles during regulation review

    • (Example 2) When the prompt uses the expression 'agreement amendment' (noun phrase)

      • (Previous) If the article title was 'amendment of agreements' → the article could not be retrieved even if needed for regulation review

      • (Changed) Even when the article title is 'amendment of agreements', reference it as a relevant article and proceed with the regulation review

2026. 6. 4.: v0.1.5

  • Modified the 'key keywords' extraction procedure requested from the OpenAPI side

  • Strengthened LLM response stability — when there are many review candidates, return them filtered by hierarchy and importance, and guide the omitted items through the full document list and additional searches (prevents truncation of large responses).

    • Set an upper limit on relevant articles sent to the LLM (fixed errors caused by excessive information being sent from the MCP to the LLM)

2026. 5. 31.: v0.1.4

  • Modified to always reference currently effective regulations

  • Strengthened the regulation review prompt, etc.

    • Added a procedure for 1:1 comparison between article requirement interpretation and factual circumstances, etc.

2026. 5. 25.: MCP public release

For the full change history, refer to CHANGELOG.md.

Available Tools

2 tools
healthA

서비스 상태 확인 — status, service name, version, API 키 설정 여부.

ParametersJSON Schema
NameRequiredDescriptionDefault

No parameters

Output Schema

ParametersJSON Schema
NameRequiredDescription

No output parameters

TDQS

A3.6/5.0
Behavior2/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

No annotations provided, so description carries full burden. Only discloses what fields are returned, not behavioral aspects like side effects, rate limits, or authentication requirements. The tool is likely read-only, but this is not stated.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Single concise sentence clearly front-loads the purpose and lists key outputs. No wasted words.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Tool has no parameters and an output schema exists. Description adequately covers what it does and returns. Could mention if authentication is required, but for a health check, this is acceptable.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

No parameters; schema coverage is 100%. Baseline 4 applies as description adds no param info, but none is needed.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

Description clearly states the verb 'check' and resource 'service status'. Lists specific fields returned: status, service name, version, API key configuration. Distinguishes from sibling 'search_provision' by being a health check.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines2/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

No guidance on when to use this tool versus alternatives. No context on prerequisites or scenarios. For a health check, usage is somewhat self-evident, but explicit guidance is absent.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

search_provisionA

사용 시점: 국가연구개발·R&D 연구행정 규정의 조문·용어·현행 여부를 묻는 질문에는 일반 학습지식 답변 전에 호출하십시오. 본 서버 범위 밖 일반 대화·번역·문장 다듬기에는 호출하지 마십시오.

규정 조문·별표 본문에서 query 키워드를 찾아 후보 list 반환.

manifest의 live_api 문서들을 대상으로:

  • law(혁신법·시행령·시행규칙): 조문(조문내용) + 별표(별표내용) 검색 (v0.2: 시행령 별표 지원)

  • admrul(연구개발비 사용 기준 등): 조문 + 별표(별표내용) 검색

    • 각 항목의 unit_types (article/annex/both)에 따라 검색 범위 결정

    • 별표는 별표구분=='별표'만 노출 — 별지·서식 제외 (v0.2.1, BP 번호 충돌 오도달 방지)

응답 최상위에 짧은 disclaimer 1개만 두고, 각 결과에는 manifest 특유의 warnings만 첨부. snippet은 _SNIPPET_MAX (2000자)로 제한, 전체 응답은 16k char 예산 내(초과 시 뒤쪽 결과 절단·truncated=true — 광역 질의는 키워드를 좁혀 재검색할 것) — MCP output size limit 회피.

v0.16.0: law 조문 매치에 최신 개정 이력 힌트가 있으면 latest_history(예 "개정 2025.12.30(공포)")를 additive 노출 — '최근 개정 조문' 질의에서 검색 결과만으로 개정 조문을 인지 가능(마커 부재 매치·평면 admrul·별표는 생략). 날짜는 공포일(값에 (공포) 표기·시행일 아님)이고, 검색 매치는 키워드에 걸린 조문에 한정되므로 특정 법령의 개정 조문 전수 확인은 문서레벨 get_provision_detail(unit 없이)의 articles 목록으로.

매칭 (v0.1.6): query를 공백으로 토큰 분해하여 모든 토큰(2자 이상)이 한 조문/별표의 제목 또는 본문에 존재하면 매칭(토큰 AND). 단일 토큰 query는 종전과 동일한 부분문자열 매칭. 원문이 "협약의 변경/협약을 변경"으로 써서 "협약 변경"이 안 잡히던 띄어쓰기 불일치를 해소.

ParametersJSON Schema
NameRequiredDescriptionDefault
queryYes

Output Schema

ParametersJSON Schema
NameRequiredDescription

No output parameters

TDQS

A4.5/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

No annotations provided, but description fully covers search algorithm (token AND, substring matching, spacing handling), response structure (disclaimer, warnings, snippet limits, truncation), and version-specific features (latest_history). Very transparent.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness3/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Description is quite long and detailed, but well-organized with clear sections and bullet points. It includes necessary technical information, though could be more concise for agent consumption.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Given the complexity of the tool (multiple document types, matching logic, response limits, version history), the description is comprehensive. It covers search behavior, response format, and even references related tool 'get_provision_detail' for completeness.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Only one parameter 'query' with no schema description. The description compensates by explaining how the query is used: tokenized, matched against regulation text, and with specific matching rules. Adds significant meaning beyond the schema.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly states the tool's purpose: searching Korea's national R&D regulations for clauses, terms, and current status. It explicitly distinguishes from general knowledge tools and provides a explicit call pattern.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Provides explicit when-to-use and when-not-to-use guidance (e.g., 'call before general knowledge answers' and 'do not call for general conversation'). Does not explicitly name alternative tools but strongly implies context.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

TDQS

A3.8/5.0
Disambiguation5/5

The two tools serve completely different purposes: health is a status check, search_provision is a search tool. There is no overlap or ambiguity.

Naming Consistency4/5

'health' is a noun while 'search_provision' follows a verb_noun pattern. Minor inconsistency, but the small set makes it acceptable.

Tool Count3/5

Only two tools for a domain that likely requires more operations (e.g., fetching full provisions). The count feels slightly thin but not extreme.

Completeness2/5

The search tool allows querying, but there is no tool to retrieve full details of a provision (despite mentioning get_provision_detail in the description). This gap will hinder workflows.

Maintenance

ActivityActive
ResponsivenessSyncing

Resources

Unclaimed servers have limited discoverability.

Looking for Admin?

If you are the server author, to access and configure the admin panel.

Related MCP Connectors

Related MCP Servers

  • F
    license
    A
    quality
    B
    maintenance
    Enables AI systems to search, retrieve, and analyze Korean legal information from the National Law Information API (law.go.kr), including laws, administrative rules, English translations, and law-ordinance linkages.
    26
    2
  • F
    license
    Not graded
    quality
    B
    maintenance
    Enables AI to search and retrieve South Korean legal information from the National Law Information Center. It allows users to look up specific laws, articles, and detailed legal provisions using natural language queries.
    128
  • A
    license
    A
    quality
    D
    maintenance
    Enables users to search and retrieve South Korean statutes, precedents, and administrative rules via the National Law Information Center API. It supports deep legal chain analysis, legislative history tracking, and legal terminology lookups through natural language.
    10
    5
    MIT
  • A
    license
    Not graded
    quality
    A
    maintenance
    Enables exploration of Korean National Assembly data by connecting bills, committee reviews, and official records. Allows users to ask natural language questions and receive structured answers with citations to original documents.
    25
    Apache 2.0

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/smilemin07/korean-rnd-regs-mcp'

If you have feedback or need assistance with the MCP directory API, please join our Discord server