LexLink-ko-mcp

🌐 Read this in other languages: English | 한국어 (Korean)

LexLink is an MCP (Model Context Protocol) server that exposes the Korean National Law Information API (open.law.go.kr) to AI agents and LLM applications. It enables AI systems to search, retrieve, and analyze Korean legal information through standardized MCP tools.

Features

• 54 MCP Tools + 2 MCP Resources for comprehensive Korean law information access

  • Search and retrieve Korean laws (effective date & announcement date)

  • Search and retrieve English-translated laws

  • Search and retrieve administrative rules (행정규칙)

  • Query specific articles, paragraphs, and sub-items

  • Law-ordinance linkage (법령-자치법규 연계)

  • Delegated law information (위임법령)

  • Phase 3 - Case Law & Legal Research

    • Court precedents (판례)

    • Constitutional Court decisions (헌재결정례)

    • Legal interpretations (법령해석례)

    • Administrative appeal decisions (행정심판례)

  • Phase 4 - Article Citation Extraction

    • Extract legal citations from any law article (100% accuracy)

  • NEW: Phase 5 - AI-Powered Search

    • Semantic search for natural language queries (aiSearch)

    • Related laws discovery (aiRltLs_search)

  • MCP Resources - Law ID Cache

    • Cached mapping of ~20 frequently-used law names to stable 법령ID codes

    • Template lookup by Korean name or abbreviation (lexlink://law/{name})

    • Dynamic caching: search results automatically populate the cache

• Smart Features (inspired by korean-law-mcp):

  • Intelligent Caching - Per-tool TTL caching (search 1hr, articles 24hr, AI search 30min)

  • Law Name Resolution - Auto-resolves Korean abbreviations (자통법→자본시장과 금융투자업에 관한 법률), 52 seed aliases + dynamic learning

  • Chain Tools - Multi-step research workflows in one call (Phase 9)

• 100% Semantic Validation - All Phase 1-5 tools confirmed returning real law data

• Error Handling - Actionable error messages with resolution hints

• Korean Text Support - Proper UTF-8 encoding for Korean characters

• Response Formats - JSON (default), HTML, or XML (multiple formats supported)

Related MCP server: Legal Search MCP

Project Status

🎉 Production Ready - Phase 9 Complete!

Latest: v2.1.0 — 54 tools (Phase 9 added), intelligent caching (cache.py), law name resolution (resolver.py), chain tools for multi-step research workflows.

Prerequisites

• Python 3.10+

• law.go.kr OC identifier: Register at open.law.go.kr

Quick Start

1. Install Dependencies

uv sync

2. Configure Your OC Identifier

Option A: Environment Variable (Recommended)

# Set OC in your environment
export OC=your_id_here

Option B: Pass in Tool Arguments

# Override OC in each tool call
eflaw_search(query="법령명", oc="your_id")

3. Run the Server

# Stdio transport (for Claude Code, Cursor, etc.)
OC=your_oc uv run stdio

# HTTP transport (for Kakao PlayMCP)
OC=your_oc TRANSPORT=http uv run serve

Available Tools

Phase 1: Core Law APIs (6 tools)

1. eflaw_search - Search Laws by Effective Date

Search for laws organized by effective date (시행일 기준).

eflaw_search(
    query="자동차관리법",      # Search keyword
    display=10,                # Results per page
    type="XML",                # Response format
    ef_yd="20240101~20241231"  # Optional date range
)

2. law_search - Search Laws by Announcement Date

Search for laws organized by announcement date (공포일 기준).

law_search(
    query="민법",
    display=10,
    type="XML"
)

3. eflaw_service - Retrieve Law Content (Effective Date)

Get full law text and articles by effective date.

IMPORTANT: For specific article queries (e.g., "제174조"), use the jo parameter. Some laws have 400+ articles and responses can exceed 1MB without jo.

# Get specific article (RECOMMENDED)
eflaw_service(
    mst="279823",              # Law MST
    jo="017400",               # Article 174 (제174조)
    type="XML"
)

# Get full law (WARNING: large response)
eflaw_service(
    id="001823",
    type="XML"
)

4. law_service - Retrieve Law Content (Announcement Date)

Get full law text and articles by announcement date.

IMPORTANT: For specific article queries (e.g., "제174조"), use the jo parameter. Some laws have 400+ articles and responses can exceed 1MB without jo.

# Get specific article (RECOMMENDED)
law_service(
    mst="279823",              # Law MST
    jo="017400",               # Article 174 (제174조)
    type="XML"
)

5. eflaw_josub - Query Article/Paragraph (Effective Date)

Best tool for querying specific articles. Returns only the requested article/paragraph.

eflaw_josub(
    mst="279823",              # Law MST
    jo="017400",               # Article 174 (제174조)
    type="XML"
)
# jo format: "XXXXXX" where first 4 digits = article (zero-padded), last 2 = branch (00=main)
# Examples: "017400" (제174조), "000300" (제3조), "001502" (제15조의2)

6. law_josub - Query Article/Paragraph (Announcement Date)

Best tool for querying specific articles. Returns only the requested article/paragraph.

law_josub(
    mst="279823",              # Law MST
    jo="017200",               # Article 172 (제172조)
    type="XML"
)

Phase 2: Extended APIs (9 tools)

7. elaw_search - Search English-Translated Laws

Search for Korean laws translated to English.

elaw_search(
    query="employment",
    display=10,
    type="XML"
)

8. elaw_service - Retrieve English Law Content

Get full English-translated law text.

elaw_service(
    id="009589",
    type="XML"
)

9. admrul_search - Search Administrative Rules

Search administrative rules (훈령, 예규, 고시, 공고, 지침).

admrul_search(
    query="학교",
    display=10,
    type="XML"
)

10. admrul_service - Retrieve Administrative Rule Content

Get full administrative rule text with annexes.

admrul_service(
    id="62505",
    type="XML"
)

11. lnkLs_search - Search Law-Ordinance Linkage

Find laws linked to local ordinances.

lnkLs_search(
    query="건축",
    display=10,
    type="XML"
)

12. lnkLsOrdJo_search - Search Ordinance Articles by Law

Find ordinance articles linked to specific law articles.

lnkLsOrdJo_search(
    knd="002118",              # Law ID
    display=10,
    type="XML"
)

13. lnkDep_search - Search Law-Ordinance Links by Ministry

Find laws linked to ordinances by government ministry.

lnkDep_search(
    org="1400000",             # Ministry code
    display=10,
    type="XML"
)

14. drlaw_search - Retrieve Law-Ordinance Linkage Statistics

Get linkage statistics table (HTML format).

drlaw_search(
    lid="001823",              # Law ID
    type="HTML"
)

15. lsDelegated_service - Retrieve Delegated Law Information

Get information about delegated laws, rules, and ordinances.

lsDelegated_service(
    id="001823",
    type="XML"
)

Phase 3: Case Law & Legal Research (8 tools - NEW!)

Tip: All *_service tools in Phase 3 support a sections="summary" parameter to return only a brief summary instead of the full document text.

16. prec_search - Search Court Precedents

Search Korean court precedents from Supreme Court and lower courts.

prec_search(
    query="담보권",
    display=10,
    type="XML",
    curt="대법원"             # Optional: Court name filter
)

17. prec_service - Retrieve Court Precedent Full Text

Get complete court precedent text with case details.

prec_service(
    id="228541",
    type="XML"
)

18. detc_search - Search Constitutional Court Decisions

Search Korean Constitutional Court decisions.

detc_search(
    query="벌금",
    display=10,
    type="XML"
)

19. detc_service - Retrieve Constitutional Court Decision Full Text

Get complete Constitutional Court decision text.

detc_service(
    id="58386",
    type="XML"
)

20. expc_search - Search Legal Interpretations

Search legal interpretation precedents issued by government agencies.

expc_search(
    query="임차",
    display=10,
    type="XML"
)

21. expc_service - Retrieve Legal Interpretation Full Text

Get complete legal interpretation text.

expc_service(
    id="334617",
    type="XML"
)

22. decc_search - Search Administrative Appeal Decisions

Search Korean administrative appeal decisions.

decc_search(
    query="*",                # Search all decisions
    display=10,
    type="XML"
)

23. decc_service - Retrieve Administrative Appeal Decision Full Text

Get complete administrative appeal decision text.

decc_service(
    id="243263",
    type="XML"
)

Phase 4: Article Citation Extraction (1 tool - NEW!)

24. article_citation - Extract Citations from Law Article

Extract all legal citations referenced by a specific law article.

# First, search for the law to get MST
eflaw_search(query="건축법")  # Returns MST: 268611

# Then extract citations
article_citation(
    mst="268611",              # Law MST from search result
    law_name="건축법",          # Law name
    article=3                  # Article number (제3조)
)

Response:

{
    "success": true,
    "law_name": "건축법",
    "article": "제3조",
    "citation_count": 12,
    "internal_count": 4,
    "external_count": 8,
    "citations": [
        {
            "type": "external",
            "target_law_name": "「국토의 계획 및 이용에 관한 법률」",
            "target_article": 56,
            "target_paragraph": 1
        }
    ]
}

Key Features:

• 100% accuracy via HTML parsing (not LLM-based)

• Zero API cost (no external LLM calls)

• ~350ms average extraction time

• Distinguishes internal vs external citations

Phase 5: AI-Powered Search (2 tools - NEW!)

25. aiSearch - AI-Powered Semantic Law Search

⭐ PREFERRED TOOL for vague or natural language queries. Use this FIRST when user's intent is unclear or conversational.

Uses intelligent/semantic search to find relevant law articles with full article text.

aiSearch(
    query="뺑소니 처벌",           # Natural language query
    search=0,                      # 0: law articles, 1: appendix, 2: admin rules, 3: admin appendix
    display=20,                    # Results per page
    page=1,                        # Page number
    type="JSON"                    # Response format (JSON default)
)

Best for: Natural language queries like "음주운전 벌금", "이혼 재산분할", "상속 문제"

26. aiRltLs_search - AI-Powered Related Laws Search

⭐ PREFERRED TOOL for discovering related laws from vague topics. Use this when user wants to explore laws around a general subject.

Finds laws semantically related to a given law name or keyword.

aiRltLs_search(
    query="민법",                  # Law name or keyword
    search=0,                      # 0: law articles, 1: admin rule articles
    type="JSON"                    # Response format (JSON default)
)

Best for: Finding related laws like "민법" → 상법, 의료법, 소송촉진법

Phase 7: Extended Legal Information (18 tools)

Phase 9: Chain Tools (5 tools)

Inspired by korean-law-mcp, these tools run multi-step research workflows in a single call — eliminating the need for an LLM to orchestrate sequential tool calls manually.

Tool Selection Guide

When searching Korean law, select tools based on query clarity:

Configuration

Environment Variables

OC Priority

When resolving the OC identifier:

1. Tool argument (highest priority) - oc parameter in tool call

2. Environment variable - OC env var (set via .env or HTTP header middleware)

Usage Examples

Example 1: Basic Search

# Search for automobile management law
result = eflaw_search(
    query="자동차관리법",
    display=5,
    type="XML"
)

# Returns:
{
    "status": "ok",
    "request_id": "uuid",
    "upstream_type": "XML",
    "data": {
        # Law search results...
    }
}

Example 2: Search with Date Range

# Find laws effective in 2024
result = eflaw_search(
    query="교통",
    ef_yd="20240101~20241231",
    type="XML"
)

Example 3: Error Handling

# Missing OC parameter
result = eflaw_search(query="test")

# Returns helpful error:
{
    "status": "error",
    "error_code": "MISSING_OC",
    "message": "OC parameter is required but not provided.",
    "hints": [
        "1. Tool argument: oc='your_value'",
        "2. Environment variable: OC=your_value"
    ]
}

Golden MCP Tool Trajectories

These examples demonstrate real-world conversation flows showing how LLMs interact with LexLink tools to answer legal research questions.

Trajectory 1: Basic Law Research

User Query: "What is Article 20 of the Civil Code?"

Tool Calls:

1. law_search(query="민법", display=50, type="XML") → Find Civil Code ID

2. law_service(id="000021", jo="002000", type="XML") → Retrieve Article 20 text

Result: LLM provides formatted explanation of Civil Code Article 20 with full legal text and context.

Trajectory 2: Court Precedent Analysis

User Query: "Find recent Supreme Court precedents about security interests"

Tool Calls:

1. prec_search(query="담보권", curt="대법원", display=50, type="XML") → Search Supreme Court precedents

2. prec_service(id="228541", type="XML") → Retrieve top precedent details

Result: LLM summarizes key precedents with case numbers, dates, and holdings related to security interests.

Trajectory 3: Cross-Phase Legal Research

User Query: "How does the Labor Standards Act handle overtime, and are there relevant court precedents?"

Tool Calls:

1. eflaw_search(query="근로기준법", display=50, type="XML") → Find Labor Standards Act

2. eflaw_service(id="001234", jo="005000", type="XML") → Retrieve Article 50 (overtime provisions)

3. prec_search(query="근로기준법 연장근로", display=30, type="XML") → Search overtime precedents

4. prec_service(id="234567", type="XML") → Retrieve leading precedent

Result: LLM provides comprehensive analysis combining statutory text with judicial interpretation, showing how courts apply the overtime provisions.

Trajectory 4: Constitutional Review

User Query: "Has the Constitutional Court reviewed laws about fines?"

Tool Calls:

1. detc_search(query="벌금", display=50, type="XML") → Search Constitutional Court decisions

2. detc_service(id="58386", type="XML") → Retrieve decision full text

3. law_search(query=<law_name_from_decision>, type="XML") → Find related law for context

Result: LLM explains Constitutional Court holdings on fine-related provisions and their impact on specific laws.

Trajectory 5: Administrative Law Research

User Query: "What administrative rules exist for schools, and are there related legal interpretations?"

Tool Calls:

1. admrul_search(query="학교", display=50, type="XML") → Search school-related administrative rules

2. admrul_service(id="62505", type="XML") → Retrieve rule content

3. expc_search(query="학교", display=30, type="XML") → Search legal interpretations

4. expc_service(id="334617", type="XML") → Retrieve interpretation details

Result: LLM provides overview of administrative framework for schools with official agency interpretations.

Trajectory 6: Comprehensive Legal Analysis

User Query: "I'm researching rental housing disputes. Show me the relevant law, court precedents, and administrative appeal decisions."

Tool Calls:

1. eflaw_search(query="주택임대차보호법", display=50, type="XML") → Find Housing Lease Protection Act

2. eflaw_service(id="002876", type="XML") → Retrieve full law text

3. prec_search(query="주택임대차", display=50, type="XML") → Search housing lease precedents

4. prec_service(id="156789", type="XML") → Retrieve key precedent

5. decc_search(query="주택임대차", display=30, type="XML") → Search administrative appeal decisions

6. decc_service(id="243263", type="XML") → Retrieve appeal decision

Result: LLM provides comprehensive legal research report covering statutory framework, judicial interpretation, and administrative precedents for rental housing disputes.

Trajectory 7: Citation Network Analysis (Phase 4)

User Query: "What laws does Article 3 of the Building Act cite?"

Tool Calls:

1. eflaw_search(query="건축법", display=50, type="XML") → Find Building Act, get MST

2. article_citation(mst="268611", law_name="건축법", article=3) → Extract all citations

Result: LLM provides complete citation analysis showing 12 citations (8 external laws, 4 internal references) including specific article and paragraph references.

Trajectory 8: AI-Powered Natural Language Search (Phase 5)

User Query: "What's the penalty for hit-and-run accidents?"

Tool Calls:

1. aiSearch(query="뺑소니 처벌", search=0, display=20, type="XML") → Semantic search for hit-and-run penalties

Result: LLM receives full article text from relevant laws (특정범죄 가중처벌 등에 관한 법률 제5조의3) with complete provisions about hit-and-run penalties, enabling comprehensive answer without needing to know specific law names.

Trajectory 9: Discovering Related Laws (Phase 5)

User Query: "What laws are related to the Civil Code?"

Tool Calls:

1. aiRltLs_search(query="민법", search=0, type="XML") → Find semantically related laws

Result: LLM discovers related laws like 상법 (Commercial Act), 의료법 (Medical Service Act), 소송촉진법 (Act on Special Cases Concerning Expedition of Litigation), showing connections across legal domains.

Key Patterns

1. AI Tools for Vague Queries: Use aiSearch or aiRltLs_search FIRST when user intent is unclear or conversational

2. Search First, Then Retrieve: Always search to find IDs before calling service tools

3. Use display=50-100 for Law Searches: Ensures exact matches are found due to relevance ranking

4. Combine Phases: Mix Phase 1 (laws), Phase 2 (administrative rules), Phase 3 (precedents), and Phase 5 (AI search) for complete research

5. Type Parameter: Default is type="JSON"; specify type="XML" if your pipeline requires XML

6. Article Numbers: Use 6-digit format (e.g., "002000" for Article 20) when querying specific articles

Development

Project Structure

lexlink-ko-mcp/
├── src/lexlink/
│   ├── server.py        # Main MCP server with 54 tools
│   ├── _helpers.py      # Shared helpers: run_search, run_service, TOOL_ANNOTATIONS
│   ├── cache.py         # Intelligent per-tool TTL caching (~183 lines)
│   ├── resolver.py      # Korean law name/abbreviation resolution (~225 lines)
│   ├── http_server.py   # HTTP/SSE server for Kakao PlayMCP
│   ├── stdio_server.py  # Stdio transport entry point
│   ├── params.py        # Parameter resolution & mapping
│   ├── validation.py    # Input validation
│   ├── parser.py        # XML parsing utilities
│   ├── ranking.py       # Relevance ranking
│   ├── citation.py      # Article citation extraction (Phase 4)
│   ├── client.py        # HTTP client for law.go.kr API
│   ├── errors.py        # Error codes & responses
│   ├── raw_logger.py    # PlayMCP traffic logging
│   └── log_processor.py # Log format converter
├── logs/playmcp/         # PlayMCP traffic logs (daily JSONL)
├── pyproject.toml        # Project configuration
└── README.md             # This file

Running Tests

# Install test dependencies
uv sync

# Run all tests
uv run pytest

# Run with coverage
uv run pytest --cov=src/lexlink --cov-report=html

# Run specific test category
uv run pytest -m unit
uv run pytest -m integration
uv run pytest -m e2e

Adding New Tools

Current Status: 54/54 tools implemented (Phase 1-9 complete). Phase 1-5 tools validated.

For implementing additional tools from the 124+ remaining APIs:

1. Follow the pattern established in src/lexlink/server.py

2. Use ctx: Context = None parameter for MCP logging/progress

3. Use generic parser functions (extract_items_list, update_items_list)

4. Add semantic validation tests

Tool Implementation Pattern:

• Each tool is a decorated function with MCP schema

• Uses ctx: Context = None parameter for MCP context

• 2-tier OC resolution: tool arg > env var

• Generic parser functions work with any XML tag

• Comprehensive error handling with actionable hints

Deployment

Deploy to Kakao PlayMCP (HTTP Server)

LexLink can also be deployed as an HTTP server for platforms like Kakao PlayMCP.

Important: Kakao PlayMCP does not accept port numbers in URLs. You must use Nginx as a reverse proxy to serve on port 80.

Quick Start (Local Testing):

# Run the HTTP server
OC=your_oc uv run serve

# Server starts at: http://localhost:8000/sse

Production Setup:

Internet → Nginx (port 80) → LexLink (port 8000)

PlayMCP Registration:

For detailed deployment instructions (AWS EC2, Nginx, systemd, HTTPS), see docs/DEPLOYMENT_GUIDE.md.

PlayMCP Traffic Logging

LexLink includes built-in logging for PlayMCP traffic analysis. Logs are saved in dashboard-compatible JSONL format.

Log Location: logs/playmcp/YYYY-MM-DD.jsonl

Log Schema:

{
  "rpc_id": "3",
  "request_id": "d8ee45eb",
  "session_id": "9ff9dc23431848a4901b4cb6326ba5bd",
  "timestamp": "2025-12-25T05:40:23.957987",
  "duration_ms": 1.52,
  "method": "tools/call",
  "tool_name": "aiSearch",
  "params": { "arguments": {"query": "뺑소니 처벌"} },
  "client": "PlayMCP",
  "client_version": "2025.0.0",
  "protocol_version": "2025-06-18",
  "client_ip": "220.64.111.219",
  "oc": "user_id",
  "status": "success",
  "status_code": 200,
  "result": { ... }
}

Features:

• Daily log rotation (one file per day)

• Dashboard-compatible format for filtering and analysis

• Captures request/response pairs with timing

• SSE streaming response parsing

Converting Old Raw Logs:

uv run python -m lexlink.log_processor input.jsonl output.jsonl

Troubleshooting

"OC parameter is required" error

Solution: Set your OC identifier using one of the three methods above.

Korean characters not displaying correctly

Solution: Ensure your terminal supports UTF-8:

export PYTHONIOENCODING=utf-8

"Timeout" errors

Solution: Increase timeout via environment variable:

export LEXLINK_TIMEOUT=90  # Increase from default 60s

Server won't start after updating dependencies

Solution: Re-sync dependencies:

uv sync --reinstall

Contributing

Contributions are welcome! Please:

1. Fork the repository

2. Create a feature branch (git checkout -b feature/amazing-feature)

3. Write tests for new functionality

4. Ensure all tests pass (uv run pytest)

5. Commit changes (git commit -m 'Add amazing feature')

6. Push to branch (git push origin feature/amazing-feature)

7. Open a Pull Request

License

This project is open source. See LICENSE file for details.

Acknowledgments

• law.go.kr - Korean National Law Information API

• MCP - Model Context Protocol by Anthropic

• korean-law-mcp - Inspiration for caching, law name resolution, and chain tools (Phase 9)

Support

• Issues: GitHub Issues

• law.go.kr API: Official Documentation

Changelog

v2.1.0 - 2026-03-30

New: Caching, Law Name Resolution, Chain Tools (Phase 9)

• Added intelligent per-tool TTL caching (cache.py): search 1hr, articles 24hr, AI search 30min

• Added law name/abbreviation resolution (resolver.py): 52 seed aliases + dynamic learning

• Added 5 Phase 9 chain tools: chain_full_research, chain_amendment_track, chain_dispute_prep, chain_law_system, cache_stats

• Inspired by korean-law-mcp

• See CHANGELOG.md for full details

v2.0.0 - 2026-03-30

Major Release: Phase 7 Tools, JSON Default, sections Parameter

• Added 18 new Phase 7 tools (자치법규, 조약, 법령정보 지식베이스, 위원회 결정문, 중앙부처 해석, 특별행정심판)

• JSON is now the default response format (was XML)

• Added sections="summary" parameter for case law service tools

• Refactored shared logic into _helpers.py

• See CHANGELOG.md for full details

v1.5.0 - 2026-02-28

Refactor: Remove Smithery Dependency

• Removed smithery package and 8 transitive dependencies

• Simplified OC resolution to 2-tier (tool arg > env var)

• Added stdio_server.py entry point for stdio transport

• See CHANGELOG.md for full details

For the full changelog (v1.0.0 – v2.1.0), see CHANGELOG.md.

Powered by MCP