Reading a P&L from a chat is common. The useful part here is the write side: approving a classification, reclassifying a vendor's charges, and turning that correction into a rule that holds next time. Each write is logged and scoped to your organisation.
A hosted MCP server over AgenticBooks, which ingests Stripe, Mercury and RevenueCat webhooks into double-entry accounting and syncs to QuickBooks or Xero. It exposes that ledger for reading and for bookkeeping actions; every write is audit-trailed and scoped to the caller's organisation.
- A P&L report for a period, and a financial summary of the business
- Live cash across every connected bank account, or one account's balance
- The review queue: unreviewed events, and what is pending grouped by counterparty
- Classifications approved, and entries reclassified when the automatic call was wrong
- Counterparty rules created and listed, so the same vendor lands correctly next month
- An audit log of what was done through the connection
An AgenticBooks account; nothing to self-host. Connect over Streamable HTTP at https://mcp.agenticbooks.ai/mcp, either through the OAuth custom-connector flow or with an ab_ API key as a bearer token. The two modes are independent — one is enough.
One command plus a key — npx -y mcp-remote https://mcp.agenticbooks.ai/mcp, then supply credentials