Fintool

What it is

Fintool was an AI equity-research copilot. You asked a question about a US-listed company in plain English and it answered out of that company's SEC filings and earnings transcripts, each claim citing the document and page it came from. It sold to hedge funds, asset managers and investment banks through a demo and a sales call, never to retail.

Microsoft acquired it on 18 April 2026 and the standalone product is gone. On 13 September 2026 the domain 301-redirects to Microsoft 365, its A records point at Azure, and api.fintool.com has no DNS record at all. The founder's farewell post — the last thing on the domain before the redirect — says the team and the IP went to Microsoft's Office Product Group, to be built into Excel, Word and PowerPoint. No shutdown date, migration path or export window was announced; the product homepage was live on 17 April 2026 and the notice replaced it that day.

Pricing

There is nothing to buy, and while there was, no number was ever published. The last live pricing page, on 15 January 2026, showed two plans and no figures: Regular, for "small to medium-sized firms", and Enterprise, which added API access, private-data integration and a dedicated support team. Both ended in a button — a free trial of unstated length, or Contact Sales. Before 12 March 2026 the page was withdrawn and /pricing began redirecting to the homepage. There was no self-serve tier and no individual plan at any point.

Data & coverage

US filers only, because the corpus was SEC-sourced: 10-K, 10-Q, 8-K, proxy and registration statements, plus earnings-call, investor-day and M&A-call transcripts, across what the vendor called 8,000 US-listed stocks. Its technology page put the pipeline at roughly two million documents and 70 million chunks, searched with BM25 and vector retrieval under a cross-encoder reranker, with XBRL parsed for statement figures rather than read off the page by a model.

The part that was not public is where the money went: an org admin could connect SharePoint and similar stores so investment memos, meeting notes and data rooms were indexed beside the filings, per-document permissions intact. Prices and computed metrics rode along and live web results came through Perplexity Sonar, but it was never a quote feed.

Integrations

An Enterprise-only REST API, v2.0, behind a bearer key issued by sales: POST /v2/chat for an answer plus a citations array of chunk id, document title and page number, and POST /v1/search for reranked chunks to feed the caller's own model. No client library was published. Reports came out as Excel, PowerPoint, Word and PDF.

Limitations

• Gone, with no successor you can buy. Microsoft's June 2026 Copilot-in-Excel finance work reaches filings through third-party connectors — Daloopa, S&P Global Kensho, FactSet — and carries neither the Fintool name nor its corpus.
• No retail route, ever. Two quote-only plans sold through a demo; a private investor could not have bought this even while it ran.
• US only, and SEC-sourced, so no LSE, Euronext, TSX or Asian filings at any price.
• The public half of the corpus was EDGAR, free at source. The subscription bought the parsing, the index and the private-document half.
• Citations located a chunk, not a number: a figure arrived with a document title and a page, which says where to look but not that the arithmetic on top of it was right. The "zero hallucination" claim and the 98% FinanceBench score were the vendor's own.
• It emitted generated prose, spreadsheets and slides — summaries, comparisons, DCF models, natural-language screens — never a published rating or price target. Every model rested on assumptions it chose, and auditing them was the reader's job.

Alternatives

Nothing replaces it whole, because half of it was your own documents. For grounded question-answering over public filings, an EDGAR MCP server puts 10-K sections and XBRL financials into any MCP client for nothing. AlphaSense is the nearest commercial equivalent and sells the same institutional way; Fiscal.ai and Koyfin are the cheaper end.

Compare it against the rest of the MCP servers and AI agents.