Japan Utils Mcp

MCP server exposing Japan-specific utilities to AI agents. Hand any Claude/Cursor/Cline agent the small bag of JP functions every Japan-related task needs but generic LLMs hallucinate on.

Why this exists

Generic LLMs are unreliable on JP-specific data:

• "What year is 令和8年?" — often wrong
• "Convert 山田太郎 to romaji" — gets the surname wrong half the time
• "What's the address for postal code 150-0001?" — fabricates plausible-looking nonsense
• "Is May 3rd a Japanese holiday?" — guesses

This MCP gives them deterministic answers backed by well-maintained Python libraries (jpholiday, posuto, pykakasi, jaconv, namedivider-python).

9 tools

• era_to_western — 令和8年 / R8 / Reiwa 8 → 2026
• western_to_era — 2026 → 令和8年 (with kanji + English + short forms)
• kanji_to_romaji — Hepburn romaji + hiragana for any Japanese text
• lookup_postal_code — 7-digit 郵便番号 → prefecture / city / area + kana
• is_holiday — date → is national holiday + JP name + weekday
• list_holidays — year → all 国民の祝日
• convert_kana — hiragana ↔ katakana ↔ half-width katakana
• normalize_width — full-width (全角) ↔ half-width (半角), per-category control
• split_japanese_name — full name → surname + given (statistical model with confidence)

Install

uvx japan-utils-mcp

PyPI: https://pypi.org/project/japan-utils-mcp/

Configuration (Claude Desktop)

{
  "mcpServers": {
    "japan-utils": {
      "command": "uvx",
      "args": ["japan-utils-mcp"]
    }
  }
}

MIT licensed. Python 3.13+. Runs locally over stdio.