Danske Bank lets business customers connect AI tools to banking data through MCP

Danske Bank launched a Model Context Protocol service for business customers on October 21, 2026, allowing approved AI tools to connect with the bank’s financial-data APIs. The bank says customers can use tools such as ChatGPT or Claude to analyze transactions, retrieve account balances and generate cash-flow forecasts.

The architecture separates discovery from live access. Danske describes a public MCP Server that lets an AI tool discover available capabilities, while actual API calls use a local MCP Proxy and require onboarding, certificates and subscriptions. That distinction matters: making an API describable is not the same as giving an AI agent permission to move or read a company’s money.

For treasury and finance teams, the potential benefit is a more natural interface to banking data. An employee could ask for unusual transactions, upcoming cash needs or a breakdown of receipts without learning a bank’s reporting menus. The usefulness will depend on data freshness, the model’s ability to cite its sources and controls that keep one customer’s information inside the correct account boundary.

The security model will be the harder part. Businesses need explicit scopes, approval workflows, audit logs, revocation and clear rules about whether an AI tool may only read data or also initiate payments. A hallucinated balance is inconvenient; an incorrectly authorized payment is an operational incident.

Danske’s launch is an early example of a bank treating AI tools as an integration surface rather than a standalone assistant. The commercial test will be whether customers gain speed without weakening identity, consent and accountability around sensitive financial data.