hike-lab/public-data-lens

View on GitHub ↗

MCP-based discovery and judgment layer for Korea’s open data portal—helping AI agents find, compare, and assess datasets with explicit evidence, JSON-LD/DCAT metadata, and versioned deterministic rules.

79 ★19 forksPythonUpdated 25d ago

What you need to know

Discovery-and-judgment MCP layer over Korea's open data portal (data.go.kr) that finds and evaluates public datasets with evidence-based reasoning.

Install

Register the connector URL https://service.datahub.kr/projects/public-data-lens/mcp in Claude web/app via Settings > Connectors
Or add {"mcpServers":{"public-data-lens":{"type":"http","url":"https://service.datahub.kr/projects/public-data-lens/mcp"}}} to .mcp.json
Self-host with docker compose after downloading the catalog CSV from data.go.kr

Usage

  • Ask in natural language, e.g. "find data needed for an elderly healthcare access analysis"
  • Use build_data_plan to get a utilization plan draft with candidates, selection rationale and expected join columns
  • Query via search_datasets, get_dataset, compare_datasets, get_catalog_changes and search_by_columns

Key features

  • Monthly catalog (~96k listings) normalized into canonical JSON-LD/DCAT with versioned deterministic quality rules
  • Deterministic tools: search_datasets, get_dataset, compare_datasets, get_catalog_changes, get_catalog_stats, search_by_columns, get_dataset_structure, build_data_plan
  • Evidence levels on every judgment (CATALOG_METADATA_ONLY, FILE_OBSERVATION, DRAFT, CANDIDATE_ONLY)
  • Reference implementation predating and validating the draft AIRD (AI-Ready Data) standard
  • Read-only free service requiring no authentication

Best for

Discovering and comparing Korean public datasets for analysis or planning with evidence.

Caveats

  • Beta public service with no SLA and a rate limit of 2 req/s per IP (20 burst)
  • No guarantee of data content, quality or joinability; final confirmation happens on the portal
  • Search queries may be included in anonymous logs; do not input personal information
Platforms: macOS · Linux · WindowsClients: Claude · ChatGPT · Claude Code
Documentation ↗

Reviewed 2026-08-11

Topics

ai-agentsai-ready-datadata-catalogdata-discoverydata-qualitydcatjson-ldkoreamcpmodel-context-protocolontologyopen-dataopen-government-datapublic-datashacl
Stars
79★
Forks
19
Language
Python
License
MIT
Created
2026-08-03
Last push
2026-08-10