Edition Api MCP Server

EDITION Intelligence Platform

The missing infrastructure for AI agents operating in Japan.

Memory API + Regulation Check API + Procedural Knowledge + MCP Server — purpose-built for Japanese business context.

---

The Problem

AI agents working with Japanese businesses hit walls that generic tools can't solve:

• Keigo (敬語): A sentence like "ワインをお持ちすれば喜ばれるかと存じます" hides the subject, uses layered honorifics, and expresses uncertainty — generic NLP treats this as noise
• Implicit agreements: Japanese business communication rarely states things directly
• Regulatory maze: 10+ industries with overlapping national/prefectural regulations, most documentation only in Japanese
• No persistent context: Agents forget everything between sessions

What This Does

1. Memory API — Japanese-aware persistent memory

Store episodes, auto-extract structured facts with keigo analysis, social hierarchy detection, and confidence scoring.

Input:  "佐藤部長にはワインをお持ちすれば喜ばれるかと存じます"

Output:
  Subject:    佐藤 (役職: 部長)
  Predicate:  好む
  Object:     ワイン
  Keigo:      Level 2 (尊敬語)
  Hierarchy:  superior
  Confidence: 0.7 (推測 — not stated as fact)
  Tense:      present

Three-layer architecture:

• Episodes — raw conversation logs
• Facts — structured knowledge (auto-extracted via LLM)
• Context — summarized state per entity/topic

2. Regulation API — 10 industries + tourist rules

Pre-built regulatory database covering:

• EC sites, Real estate, Staffing, Food service, Construction
• Healthcare, Finance, Transport, Education, Accommodation
• Tourist categories: Visa, Tax-free, Transit, Medical, Manners

All 10 industries include step-by-step procedural guides (65 total steps) — covering what to do, how, where, required documents, costs, timelines, and common pitfalls.

curl -X POST /api/v1/regulation/check \
  -d '{"industry": "food_service", "query": "What licenses do I need to open a restaurant in Tokyo?"}'

3. MCP Server — 8 tools for Claude, Cursor, etc.

Quick Start

Backend

git clone https://github.com/hiroshic9-png/edition.git
cd edition
python3 -m venv venv && source venv/bin/activate
pip install fastapi 'uvicorn[standard]' pydantic sqlalchemy aiosqlite chromadb python-dotenv google-genai

# Set your LLM key (any one of these)
echo 'GEMINI_API_KEY=your_key' > .env
# or ANTHROPIC_API_KEY or OPENAI_API_KEY

python -m uvicorn backend.api.main:app --reload
# → http://localhost:8000/docs

MCP Server (for Claude Desktop / Cursor)

cd mcp-server && npm install && npm run build && npm start

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "edition": {
      "command": "node",
      "args": ["/path/to/mcp-server/dist/index.js"],
      "env": {
        "EDITION_API_URL": "http://localhost:8000",
        "EDITION_API_KEY": "your_api_key"
      }
    }
  }
}

API Endpoints

Memory

Regulation

Tech Stack

Why Not Mem0 / Letta / Zep?

Those are excellent general-purpose memory tools. But they don't:

• Parse Japanese keigo levels (丁寧語 / 尊敬語 / 謙譲語)
• Detect implicit social hierarchy from honorific patterns
• Score confidence based on Japanese speech patterns (断定 vs 推測 vs 伝聞)
• Include a Japanese regulatory database

This project exists because Japanese business context is structurally different, and agents need purpose-built infrastructure to navigate it.

License

MIT