Model Context Protocol

Model Context Protocol (MCP) is an open standard, originally developed by Anthropic, that gives AI assistants a consistent way to connect to external systems — accounting platforms, databases, APIs — and interact with them using structured calls rather than freeform chat. Instead of pasting a spreadsheet into a chat window, the AI reaches directly into the connected system, reads the data it needs, and can write results back. Authentication, permissions, and available actions are declared by the server exposing the integration.

MCP and Xero

Xero publishes an official MCP server covering your organisation’s data. An AI assistant connected to it can retrieve balances, list open invoices, and create draft transactions — no file exports needed.

The important limit is that Xero’s MCP server cannot match and reconcile bank statement lines. Reconciliation — marking a statement line as matched against an invoice or spend money transaction — sits inside Xero’s core engine and is not exposed through the Accounting API or the MCP layer. Xero’s own JAX feature handles high-confidence matches automatically; the remainder need a human decision inside the Xero interface.

In practice, MCP makes an AI assistant far more useful for querying, drafting, and reporting, while the actual reconciliation step remains either JAX-automated or manually confirmed in Xero.