본문 바로가기
  • Home

NL2MCP: A New Framework for Natural Language Query Execution

  • Journal of The Korea Society of Computer and Information
  • Abbr : JKSCI
  • 2025, 30(10), pp.43~52
  • Publisher : The Korean Society Of Computer And Information
  • Research Area : Engineering > Computer Science
  • Received : September 22, 2025
  • Accepted : October 20, 2025
  • Published : October 31, 2025

Won-Bae Kim 1 Nammee Moon 1

1호서대학교

Accredited

ABSTRACT

This study compares two approaches for natural language query execution, NL2SQL and NL2MCP, under identical conditions. Using GPT-4o, PostgreSQL, and the Spider dev dataset, we evaluated three paths: NL2SQL(cache), NL2MCP(cache), and NL2MCP(live). Metrics included exact match (EM), execution match (EX), mean/p95 latency, and error rates across difficulty levels (easy, medium, hard, extra). Results show that NL2MCP yielded lower EM but higher EX and more stable latency. In particular, the live mode reduced schema mismatch errors and achieved the highest success rate in hard/extra queries, indicating that NL2MCP(live) is the most reliable path in real-world environments.

Citation status

* References for papers published after 2024 are currently being built.