Japan Business Data

AI-citable Japanese public business intelligence for all AI agents — 6-source Authority Chain: corporate registries (gBizINFO + 国税庁 法人番号 公表 Web-API), government subsidies (J-Grants), listed-company IR (EDINET 金融商品取引法開示書類), statutory law (e-Gov 法令検索, デジタル庁), and government procurement (官公需情報ポータル, 中小企業庁) with 士業 + investment-disclaimer boundary safety. For AI builders (Claude / Cursor / Codex / OpenAI App Server / all MCP clients), Japanese 士業 (行政書士 / 中小企業診断士 / 公認会計士 / 税理士 / 弁護士 / 司法書士), investment professionals, M&A advisors, and B2G sales / consulting teams.

This is the Japan-market equivalent of Anthropic's financial-services agent templates, focused on Japanese public canonical data sources. It is open source under Apache License 2.0 and packages Kokai Data (an AI-citable public business intelligence MCP server) as both a universal MCP server (for all AI clients) and Claude plugins + managed agents (for Anthropic ecosystem sugar).

Architecture — Universal MCP server + Anthropic Plugin sugar (2 layers)

Kokai Data follows a 2-layer architecture for AI agent ecosystem neutrality:

Layer	What	Who can use it
L1: Universal MCP server	https://mcp.kokai.ai/functions/v1/mcp-server — implements MCP spec 2025-06-18	All MCP clients: Claude Code / Claude Cowork / Claude Managed Agents / Claude.ai (connectors layer) / Cursor / Codex / OpenAI App Server / any custom MCP client. Setup via .mcp.json config (no Plugin required).
L2: Anthropic Plugin sugar (this repo)	Kokai-Data/japan-business-data — bundles Skills, slash commands, managed-agent cookbook, and auto-registers the MCP server	Anthropic ecosystem users: Claude Code (CLI + Desktop App) / Claude Cowork / Claude Managed Agents. Plugin install auto-configures everything (no manual .mcp.json editing).

Non-Claude clients (Cursor / Codex / etc.) do NOT need to install this Plugin — they connect to the MCP server directly via their own .mcp.json (see Path 4 below). The Plugin layer is purely Anthropic-ecosystem distribution sugar (auto-install / auto-update / Skills auto-invocation).

This dual approach matches Kokai's brand promise: AI agent ecosystem neutral, not tied to any single AI provider.

Naming

Brand	Kokai Data
Marketplace handle	kokai
GitHub repository	Kokai-Data/japan-business-data
Plugin (recommended install)	kokai-data
MCP server	mcp.kokai.ai
Trade sub-brand	Kokai Trade

The recommended install is the kokai-data plugin (/plugin install kokai-data@kokai) — the brand-bearing core plugin that bundles the MCP server, slash commands, and skills. The marketplace handle is kokai (brand-first); the GitHub repo stays Kokai-Data/japan-business-data.

Migration: agent plugins consolidated into kokai-data (v0.4.0)

v0.4.0 から、7 本の agent plugins (proposal-prep-jp / subsidy-fit-jp / due-diligence-jp / competitor-brief-jp / subsidy-landscape-jp / legal-research-jp / procurement-discovery-jp) は kokai-data に統合されました。機能はすべて kokai-data 1 本に含まれます（agents / slash commands / skills / MCP 接続）。

If you previously installed any of the 7 agent plugins:

1. Uninstall the agent plugin(s) (claude plugin uninstall proposal-prep-jp@kokai etc., or remove in the GUI plugin list).
2. Install / update the single plugin: claude plugin install kokai-data@kokai.
3. Your slash commands (/提案準備, /企業調査, /proposal-prep, etc.) and agents work the same — the agent name prefix changes from proposal-prep-jp: to kokai-data: when invoked as a subagent type.

Why: 8 entries in the marketplace list confused first-time users ("which do I install?"). One plugin = one instruction: install kokai-data, get everything.

Migration from legacy plugin name

The core plugin was renamed from japan-public-business-intelligence to kokai-data (for brand visibility in the Claude Code plugin list). If you installed the legacy plugin, please re-install:

1. Uninstall the legacy plugin japan-public-business-intelligence (or remove it in the GUI plugin list).
2. Install the new plugin: /plugin install kokai-data@kokai.

Your existing slash commands (/企業調査, /提案準備, etc.) work the same after re-install.

Disclaimer

Kokai provides signal / 確認材料 / context, not decisions. Subsidy 適格性, 申請可否, legal interpretation, 訴訟戦略, bid eligibility, tax, or professional judgments are out of scope and must be verified by a certified Japanese 士業 (registered advisor: 行政書士 / 中小企業診断士 / 公認会計士 / 税理士 / 弁護士 / 司法書士). Never fictionalize company, subsidy, statute, or procurement facts — use only Kokai tool/prompt outputs and cited public sources.

Who is this for

This Plugin is for:

• AI agents / AI builders (Claude Code / Claude Cowork / Cursor / Codex / OpenAI App Server) needing AI-citable Japanese public business intelligence with verifiable government sources for citation-required outputs.
• Japanese 士業 (行政書士 / 中小企業診断士 / 公認会計士 / 税理士 / 弁護士 / 司法書士) preparing meeting briefs, due diligence, competitor research, subsidy fit analysis, legal research / 法令調査, and B2G procurement discovery — with automatic 士業 boundary disclaimers that prevent AI from rendering professional judgments.
• Investment professionals / M&A advisors doing Japanese target diligence with registry change history (社名変更 / 移転 / 合併 / 解散) from 国税庁 公式 master + 上場企業 IR (EDINET 有価証券報告書 / TOB / 大量保有) + applicable 業法 verification (e-Gov 法令検索).
• B2G sales / consulting teams identifying fresh government procurement opportunities (Call for Tender / 公告) by industry + region from 官公需情報ポータル.
• Government / public sector developers building citizen-facing AI tools requiring official source attribution (cite_required + signal-not-decision layer).

This Plugin is NOT for:

• Real-time market data or pricing (use Bloomberg / Refinitiv, or anthropics/financial-services for Western markets).
• Non-Japanese companies (use anthropics/financial-services for that).
• Final eligibility / legal / tax / professional judgments — these always require a certified Japanese 士業.
• Procurement bid outcomes / 落札結果 / 受注実績 / 価格 signals (NOT in 官公需 official API; consult commercial DB like 帝国データバンク 入札情報 or hire a 中小企業診断士).

What's in this repo

Category	Count	Examples
Plugin (all-in-one)	1	kokai-data — MCP connector + 19 skills + 16 slash commands (JA/EN) + 7 agents (proposal-prep / subsidy-fit / due-diligence / competitor-brief / subsidy-landscape / legal-research / procurement-discovery)
Managed-agent cookbook (Claude Managed Agents deployment template)	1	kokai-due-diligence-jp

v0.4.0 で agent plugins 7 本を kokai-data に統合しました（一覧に並ぶ plugin は 1 本だけ。install も 1 コマンドだけ）。旧 agent plugin をお使いの方は下の Migration を参照。

Agents (7 named templates — kokai-data に同梱)

Agent	Function	Authority source
proposal-prep-jp	1-page brief for a Japanese company before a meeting (cited public data)	gBizINFO + 国税庁
subsidy-fit-jp	3-axis fit signal (regional / industry / scale) for a Japanese company × subsidy pair	gBizINFO + J-Grants
due-diligence-jp	1-page due diligence brief on a Japanese company (gBizINFO entity + evidence refs + EDINET IR)	gBizINFO + 国税庁 + EDINET
competitor-brief-jp	3-tier competitor brief (overview / scale / certifications) for a Japanese company	gBizINFO + 国税庁
subsidy-landscape-jp	Top-N Japanese subsidies for an industry / use-purpose theme (landscape scan)	J-Grants
legal-research-jp	Japanese statute lookup + verbatim 条文 fetch from e-Gov 法令検索	e-Gov 法令 (デジタル庁)
procurement-discovery-jp	Japanese government procurement (Call for Tender) discovery (Phase 1 = notice discovery only)	官公需情報ポータル (中小企業庁)

日本語 slash commands (日本人ユーザー向け alias)

英語 slash commands は日本人ユーザーには意味が把握しにくいため、日本語 alias を並列追加 (OSS rev 0.7.0)。両方が動作 (後方互換維持、英語 command も touch せず)。

日本語	英語 alias	機能
/提案準備	/proposal-prep	提案前 1 ページ ブリーフ生成 (gBizINFO + 国税庁 + J-Grants landscape)
/企業調査	/due-diligence	1 ページ デューデリ brief (gBizINFO + 国税庁 + EDINET 上場時)
/補助金マッチ	/subsidy-fit	3 軸 (地域 / 業種 / 規模) fit signal (Kokai は判定者ではなく signal 提示者)
/競合調査	/competitor-brief	3-tier 競合 brief (概観 / 規模 / 認定)
/補助金スキャン	/subsidy-landscape	Top-N 補助金 landscape scan (signal overlay 形式)
/法令調査	/legal-research	日本法令 keyword 検索 + 条文 verbatim 取得 (e-Gov, デジタル庁)
/調達公告検索	/procurement-discovery	政府調達 公告 discovery (Phase 1 = CFT only, 官公需情報ポータル)

自然言語でも auto-invoke 可能 (例: 「提案準備して」「企業調査お願い」「○○ 法人の経営情報取得」「医薬品関連の法令を調査して」「システム開発の公共調達を検索」)。日本語 keyword は agent.md triggers (7 件、各 5-7 keyword 追加) と vertical + agent SKILL.md description (59 mirror、19 unique skill) に組み込み済。

士業セーフティ表現

全 14 alias は「判定 / 決定」表現を避け、「準備 / 調査 / マッチ / スキャン / 検索」phrasing で統一。Kokai は判定者ではなく signal / 確認材料 提示者、士業 (中小企業診断士 / 行政書士 / 公認会計士 / 税理士 / 弁護士 / 司法書士) 確認前提。

Use cases (concrete examples)

1. Proposal preparation in 30 seconds

"Tell me about 株式会社サンプル製造 (法人番号 1234567890123) before my meeting"

→ The proposal-prep-jp agent invokes gbizinfo-entity-lookup + nta-corporate-number-lookup and produces a 1-page brief with cited financials, employees, certifications, registry change history, and a 4-layer authority strip + 士業 boundary disclaimer.

2. M&A due diligence with registry change history + 上場企業 IR

"Run due diligence on 法人番号 7000012050002 including past name changes, mergers, and any EDINET filings"

→ The due-diligence-jp agent combines gBizINFO financial profile + 国税庁 公式 異動履歴 (社名変更 / 移転 / 合併 / 解散 events) + EDINET 有価証券報告書 / 大量保有 / TOB disclosures + evidence refs + 4-layer authority strip. Useful when raw gBizINFO data omits change history that 国税庁 公式 master uniquely provides, or when target is listed and IR depth matters.

3. Subsidy fit signal (regional × industry × scale)

"Is the 中小企業庁 IT 導入補助金 a fit for 株式会社サンプル製造?"

→ The subsidy-fit-jp agent combines gBizINFO company profile + J-Grants subsidy detail → 3-axis fit signal (regional / industry / scale). Always includes 士業 boundary: final 適格性 / 申請可否 verification requires a certified 行政書士 or 中小企業診断士 — Kokai itself never renders eligibility decisions.

4. Legal research — verbatim 条文 + 業法 mapping for DD

"What does Article 15 of the 個人情報保護法 say? Also, list all 業法 applicable to a 医薬品 manufacturer"

→ The legal-research-jp agent invokes search-egov-laws (keyword search) + get-egov-laws-data (full statute fetch) from e-Gov 法令検索 (デジタル庁). Output is verbatim 公式 statute text with law_id + enforcement_date + e-Gov URL citation. Never legal interpretation — final 法的判断 / 訴訟戦略 requires certified Japanese 弁護士 / 行政書士 / 司法書士.

5. B2G sales lead discovery — government procurement notices

"Show me recent システム開発 procurement notices issued by 中小企業庁 or 東京都"

→ The procurement-discovery-jp agent invokes search-procurement-portal from 官公需情報ポータル (中小企業庁). Returns CFT notices with project_name / organization_name / category / procedure_type / cft_issue_date / source_url. Phase 1 scope: discovery only. Phase 2 (受注実績 / 落札結果 / 落札価格 / supplier 履歴) NOT in official API — direct user to commercial DB or 中小企業診断士 consultation for outcomes.

Repository layout

japan-business-data/
├── .claude-plugin/
│   └── marketplace.json
├── plugins/
│   └── vertical-plugins/
│       └── kokai-data/                    # ← the ONE plugin (v0.3.0 で agent plugins を統合)
│           ├── agents/                    # 7 agents (proposal-prep / due-diligence / subsidy-fit / competitor-brief / subsidy-landscape / legal-research / procurement-discovery)
│           ├── commands/                  # 16 slash commands (日本語 8 + English alias 8)
│           └── skills/                    # 19 shared skills
├── managed-agent-cookbooks/
│   └── kokai-due-diligence-jp/
├── scripts/
├── README.md
├── CHANGELOG.md
├── LICENSE
├── TRADEMARKS.md
└── NOTICE

Getting started — 4 paths

Path 1A: Claude Code (CLI)

claude plugin marketplace add Kokai-Data/japan-business-data
claude plugin install kokai-data@kokai   # これ 1 本で全部入り (MCP + slash + skills + agents)

Path 1B: Claude Code Desktop App (Windows / macOS, no CLI required)

1. Open Claude Code Desktop App → sidebar Customize
2. Under 個人用プラグイン (Personal plugins) click the + button
3. Select プラグインを作成 → マーケットプレイスを追加 (Create plugin → Add marketplace)
4. Enter repository: Kokai-Data/japan-business-data (or full URL https://github.com/Kokai-Data/japan-business-data)
5. After the marketplace is added, install kokai-data — the single all-in-one plugin (MCP + slash commands + skills + 7 agents). Nothing else to choose.

This is the most user-friendly path for users not comfortable with CLI. The Plugin auto-registers the kokai MCP server, so no .mcp.json editing is required.

Path 2: Claude Cowork (UI)

1. Open Claude Cowork plugin manager
2. Add this repo URL (this plugin is intended for submission to the Claude plugin marketplace)
3. Enable the agents / verticals you need

Path 3: Claude Managed Agents (API)

export ANTHROPIC_API_KEY=sk-ant-...
scripts/deploy-managed-agent.sh kokai-due-diligence-jp

Path 4: Non-Anthropic MCP clients — direct MCP server connection (no Plugin install)

If you use Cursor, Codex, OpenAI App Server, or any other MCP-compatible client, you don't need to install this Plugin. Connect directly to the kokai MCP server via your client's MCP config (.mcp.json or equivalent):

{
  "mcpServers": {
    "kokai": {
      "url": "https://mcp.kokai.ai/functions/v1/mcp-server"
    }
  }
}

The MCP server exposes all kokai tools (gBizINFO / J-Grants / 国税庁 法人番号 / EDINET / e-Gov 法令検索 / 官公需情報ポータル) directly. You won't get the Skills / slash commands bundled in this Plugin, but you get full data access via the MCP tools (6-source Authority Chain + signal overlay).

Skills + slash commands (e.g., /proposal-prep, /due-diligence, /legal-research, /procurement-discovery) are Anthropic-ecosystem-specific sugar bundled in this Plugin — they're convenience layers for Claude Code / Cowork / Managed Agents, not part of the MCP protocol. Non-Anthropic clients build their own workflows on top of the raw MCP tools.

How it fits together

• Agents are workflows that own end-to-end processes (e.g., proposal prep, legal research, procurement discovery).
• Skills are reusable expertise (e.g., gbizinfo-entity-lookup, search-egov-laws, search-procurement-portal) living in the vertical plugin, bundled into agents.
• Commands are explicitly-triggered slash commands (e.g., /proposal-prep, /legal-research, /procurement-discovery).
• MCP servers (declared in .mcp.json) wire to external data; here, the kokai MCP server provides Japan public business intelligence.
• Managed-agent cookbooks package the same skills as deployable templates for Claude Managed Agents API.

Vertical plugin

Plugin	Focus	MCP server
kokai-data	Japanese corporate registries, subsidies, listed-company IR, statutory law, government procurement, evidence citations	kokai

MCP integrations

MCP server	URL	Source
kokai	https://mcp.kokai.ai/functions/v1/mcp-server	gBizINFO + J-Grants + 国税庁 法人番号 + EDINET 金融商品取引法開示書類 + e-Gov 法令検索 (デジタル庁) + 官公需情報ポータル (中小企業庁)

6-Source Authority Chain

1. gBizINFO (経済産業省) — company business intelligence (financials / employees / certifications / shareholders).
2. J-Grants (中小企業庁 + デジタル庁) — government subsidies (eligibility / deadlines / max amount / application forms).
3. 国税庁 法人番号 公表 Web-API — authoritative corporate-number master + change history (社名変更 / 移転 / 合併 / 解散).
4. EDINET (金融庁) — listed-company IR (有価証券報告書 / 四半期 / 半期 / 臨時 / 大量保有 / 公開買付 / 内部統制 等) + Kokai signal cluster overlay (annual / quarterly / large-shareholding / TOB / governance / event / correction).
5. e-Gov 法令検索 (デジタル庁) — Japanese statutory law (Constitution / Acts / Cabinet Orders / Ministerial Ordinances / Rules) with full statute text + chapter / article structure.
6. 官公需情報ポータル (中小企業庁) — government procurement (Call for Tender notices, 公告) — project names / issuing organizations / industry categories / bid procedure types / CFT issue dates. Phase 1 = discovery only; awarded results deferred to Phase 2.

→ AI agent が引用する際に「6 公式 SoT 統合 + AI-citable signal overlay」と明示可能。Japan 公的データの法人特定 + 補助金 + 異動履歴 + 上場企業 IR + 統治法令 + 政府調達公告まで 1 接続で full coverage、AI HYVE / N-3 (3-source 競合) や Anthropic finance-agents (西洋金融 vertical) との明確 differentiator。

Boundary (士業 safety, REQUIRED)

All outputs include a 4-layer authority strip:

1. 公式 (official) — cite_required
2. Kokai normalized — cite_required
3. AI summary
4. AI estimate

Subsidy 適格性, 申請可否, 法令解釈 / 訴訟戦略, 入札判断 / 適格性判断, legal, tax, or professional judgments are out of scope. Final verification must be done by a certified Japanese 士業 (行政書士 / 中小企業診断士 / 公認会計士 / 税理士 / 弁護士 / 司法書士).

Customization

Each agent / vertical plugin is a directory with markdown skills and JSON manifests. Fork this repo, edit the markdown files, and re-install with claude plugin install. See Anthropic's plugin documentation for details.

Contributing

1. Fork the repo
2. Run python3 scripts/check.py to lint manifests and verify cross-file references
3. Open a PR

Changelog

See CHANGELOG.md for Sprint-by-Sprint history (Sprint 14+8 EDINET / 14+12 e-Gov 法令 / 14+13 官公需 etc.).

License

Apache License 2.0 — see LICENSE.

See TRADEMARKS.md for trademark notice and third-party brand allowlist policy.