Japan Finsight: Every EDINET disclosure, structured for AI

Ask Japan's disclosure record anything.

Connect Claude, ChatGPT, or any MCP client. Every EDINET filing from all 3,800+ listed companies, structured in English, every number linked to its source.

$99/mo·14-day refund window·Up and running in minutes. We'll help you get connected.

Built by the author of edinet-tools, the open-source library behind the data.

EDINET coverage

The complete record

Whatever you ask about a company, the filing is already here: every major disclosure type on EDINET, the FSA's official system, for every listed company.

EDINET disclosure	What your AI gets back	History
Annual securities reports	Audited financials, 60+ typed fields, segments, every listed company	since 2018
5%+ shareholder disclosures	Holder, stake %, joint filers untangled, stated intent	since 2021
Quarterly reports	Historical financials, through the 2024 FSA reform	2018–2024
Semi-annual reports	Half-year results	since 2021
Extraordinary reports	M&A, restructurings, governance, tagged by event	since 2024
Treasury-stock acquisitions	Buyback size and execution	since 2025
Tender offers	Registration, target opinion, final result	since 2020
Shelf registrations	Issuance plans before the raise	since 2024
Internal-control reports	J-SOX status	since 2021

Signals

Catch the catalysts, not just the financials.

The signals beyond the financials: who's accumulating, what they want, how capital gets returned.

Activist intent What each filer wants, in their own words (buybacks, board changes, spin-offs, take-privates), flagged when it's a known activist.

Capital policy The payout target when the company states one, an honest “not stated” when it doesn't.

Material events M&A, capital returns, management changes, and governance shifts, tagged and searchable.

Related-party transactions Counterparty, type, amount, terms. Source quote per item.

Cross-shareholdings 5%+ holders, plus smaller structural holdings from annual reports.

Ownership timelines Who entered, who increased, who exited, when, with stakes.

Query

Screen all 3,800+ listed companies in one query.

Combine signals from different disclosures (a buyback, a 5%+ stake, a cheap multiple) into a single screen. run_sql lets your AI query the whole record directly, surfacing names you wouldn't have thought to look for.

your agent might write one query across three disclosure types:
SELECT DISTINCT s.ticker, s.roe
FROM securities_reports s                              -- annual report
JOIN treasury_stock_reports b ON b.edinet_code = s.edinet_code        -- buyback · Doc 220
JOIN large_holding_reports h ON h.target_edinet_code = s.edinet_code  -- 5%+ stake · Doc 350
WHERE s.roe > 0.15 AND h.is_known_activist_filer;

Recent updates →

Open-source parsers, the same code that runs in production.

Underneath it all is edinet-tools, the open-source EDINET library. MIT-licensed, on PyPI for over two years, and the same parser that runs here. The code behind your numbers is public, so you can read exactly how each value is pulled.

On top, Japan Finsight adds the parts that take judgment: activist intent, capital policy, tagged events, cross-shareholdings.

XBRL-first: no OCR or LLM in the numbers. Consolidated, not parent. Every value source-linked to its disclosure. See how the data is built →

Matt Helmer, author of edinet-tools.

Japan Finsight is an independent product. Not affiliated with the Financial Services Agency, EDINET, or any disclosed company.

Ask Japan's disclosure record anything.

Japanese companies are buying back stock, unwinding cross-holdings, and drawing more activist filings.

Go wide across the market and deep on any name.

The complete record

Catch the catalysts, not just the financials.

Screen all 3,800+ listed companies in one query.

Open-source parsers, the same code that runs in production.

One plan. Every company, every major disclosure, in the AI you already use.