dqj1998/japan-company-info-mcp-bridge

The only **100% OFFLINE**, all-in-one Japanese Corporate Due Diligence MCP. Integrates **4 official
government sources**: NTA 13-digit Corporate IDs, EDINET XBRL financials (US GAAP / IFRS mapped),
e-Stat industry benchmarks, and gBizINFO certifications. **Zero cloud leakage. Zero Docker.**
Query EDINET statutory filings (5-year XBRL), National Tax Agency 13-digit corporate numbers,
major shareholders, and gBizINFO certifications/subsidies — with a built-in
**J-GAAP / US GAAP / IFRS** mapping dictionary that correctly distinguishes
営業利益 (Operating Income) from 経常利益 (Ordinary Income, a J-GAAP-only concept).
This package is the **Free edition**: **192 large-cap blue-chip listed companies** (EDINET-listed,
Nikkei 225 sample). The full dataset — **3,193+ listed companies** with complete XBRL financials,
major shareholders, gBizINFO details, and 財務省 法人企業統計調査 industry benchmarks — is available
as a one-time purchase at **mcporb.store**.
> **Use cases:** cross-border equity analysis, KYB / due-diligence entity verification,
> M&A target screening, and reading Japanese filings in English without mistranslating
> 営業利益 (Operating Income) vs 経常利益 (Ordinary Income, a J-GAAP-only concept).
---
How it works
The server bundles the mcporb-runtime binary and a pre-indexed Orb (BM25 + trigram +
optional dense-vector retrieval). **Retrieval runs on your machine.** On first launch the
runtime downloads its query-embedding model (~220MB) in the background to enable the
semantic vector method; until that finishes — and forever after, offline — the bm25,
trigram, and auto methods work without any network access. Your queries are not sent to
a third-party API by this server.
It exposes five domain tools that normalize your request into a precise query, plus the
generic search_knowledge tool as a fallback for open-ended questions:
- **edinet_financials_usgaap(company_name, metric?, fiscal_year?)** — EDINET statutory
financials (P/L, balance sheet, cash flow) and executive compensation, with accounting
concepts mapped across J-GAAP / US GAAP / IFRS.
- **japan_corporate_registry(company_name, info_type?)** — 13-digit National Tax Agency
corporate number, registered address, legal status, and gBizINFO certifications / subsidies.
- **japan_shareholders(company_name, top_n?)** — major shareholders and ownership
structure from the 大株主 section of 有価証券報告書.
- **japan_industry_benchmarks(industry, metric?)** — sector benchmarks (operating margin,
ordinary margin, equity ratio, ROE) from 財務省 法人企業統計調査 via e-Stat.
- **japan_company_search(query, method?, top_k?)** — keyword / fuzzy / semantic search to
discover a company when the target is unknown or ambiguous.
- **search_knowledge(query, method?, top_k?)** — raw knowledge-base search (fallback).
method ∈ auto (default) · bm25 (exact keyword) · trigram (fuzzy / identifier) ·
vector (semantic) · hybrid (RRF fusion).
The five domain tools resolve into search_knowledge internally, so retrieval and the
.orb capsule stay completely generic.
---
Quick start
Claude Desktop
Add to claude_desktop_config.json:
(Before the package is published to npm, use the GitHub form: