Documentation

Connect your AI to FinStat

FinStat is an accounting engine your AI connects to over the Model Context Protocol. Give it actual source documents — bank and card statements, payout reports, invoices, receipts — and it returns reconciled, review-ready double-entry books, every number traceable to the page it came from.

Connecting

Add the server URL to any MCP client — Claude, an IDE, or your own agent. The client discovers auth.finstat.ai through standard OAuth protected-resource metadata, registers itself, and walks you through sign-in. There are no API keys to copy and nothing to configure by hand.

The workflow

FinStat is documents-first. The canonical path, which the get_started tool returns to your AI verbatim:

1. Create or recover a workspace — create_workspace. The workspace scopes documents, shared reality, optional charts, and reports.
2. Turn sources into Standard Documents — create_standard_statements reads a PDF and produces sealed, portable Standard Statements as a durable Task. Keep the files; any FinStat system can reproduce the books from them.
3. Submit — one atomic batch. FinStat validates everything before admitting anything; a document that does not reconcile is refused with the reason named.
4. Ask for the books— balance sheet, P&L, trial balance, events, parties, account activity. A chart of accounts is optional: a chart is a model, and your books are a projection of the transaction data onto it. One workspace can carry several.

Anything ambiguous comes back as a review item for you to decide — never quietly guessed.

Metering

Every tool description states whether it is metered. Reads, listings, reports, verification, rendering, and exports are free. Model-backed interpretation — turning a PDF into Standard Statements, authoring a chart or declaration — is metered, and exact-byte replays of already-interpreted sources are not charged again.

Generated from the live tool surface 62644355c35b · 51 tools · MCP 2026-07-28 / 2025-11-25 / 2025-06-18 / 2025-03-26