Finologee has built a native Model Context Protocol (MCP) server into Banking Orchestrator. For clients, it means the platform can now be reached from any AI assistant that supports the MCP standard, Claude and ChatGPT among them. The AI assistant can ask the Orchestrator questions and carry out actions, subject to the same controls that govern the platform today.

What changes in practice

Banking Orchestrator is one platform for every bank, with multi-bank visibility, payments and liquidity automation. The MCP integration adds another way to work with it, straight from the AI assistant you already use: a request reads from or acts on the platform directly, without opening the portal. Within the rights of the authenticated user, that covers:

  • Reading: payment history and status, counterparty and signatory details, approval status and audit-ready exports.
  • Acting: initiate payments and SEPA direct debit collections, run account sweeps and update counterparty records.
Possible use cases

Our latest release lays the groundwork for working with Banking Orchestrator in plain language. Ask it what is waiting on a second signature this morning, or which large payments were approved last quarter and by whom, and get the answer in seconds, with no export and no support ticket. It can also carry out the work, initiating a batch of SEPA collections or running account sweeps from a single request, always subject to validation by the signatories. Because the AI assistant reasons over live platform data, it can go further than reading and acting on the present:

  • Forecasting: working from the payment history across several accounts, it reads the patterns in incoming and outgoing flows, recurring collections, supplier cycles, payroll and capital calls, and projects them forward. The result is a view of the coming week’s or month’s cash position across every bank, with expected shortfalls and surpluses identified before they arrive, so funding and redeployment decisions are taken ahead of value date rather than after it.
  • Anomaly detection: given a set of accounts, it examines transactions for what does not fit, such as amounts outside the usual range, unfamiliar counterparties, payments that break an established cadence or signatories acting outside their normal pattern. This turns an occasional, sampled check into a systematic one and surfaces the outliers that warrant a closer look, whether the cause is fraud or simple error.
  • Compliance: across a defined range of accounts, it crosses the statutory signing rights held for each entity against the signatures actually applied and reports where the two diverge, such as a payment released below the required number of approvers, a mandate that no longer matches the account or an authorisation given outside a signatory’s remit. Work that means reconciling mandates against bank records by hand becomes a single request.
Same controls, every action

It is a small change on the surface with a large implication: where most AI assistant integrations stop at reading data, Finologee’s Banking Orchestrator lets the AI assistant act too, with every action passing through the same controls as any other action on the platform:

  • It goes through authentication, user rights and scope, the multi-party signatory and four-eyes approval engine.
  • Write actions are designed to require an explicit confirmation before anything is executed, so nothing runs in the background.
  • It is recorded in the audit log under the name of the user who triggered it. If someone uses an AI assistant to submit a payment, the platform records it as created by that person, not by an anonymous AI assistant.
  • It stays within the user’s permissions: the AI assistant cannot suppress a signatory requirement, override a blacklist entry or reach beyond the configured scope.

The Orchestrator was built so that compliance controls sit inside the workflow rather than alongside it. The MCP integration extends that reach to wherever the user is working, without relaxing any of it.

“Security was the starting point, not an afterthought. The MCP server runs on the same authentication and permission model as the rest of Banking Orchestrator, so an AI assistant acts only as the signed-in user and only within that user’s rights. Nothing runs anonymously and nothing sits outside the platform’s controls.”

Georges Berscheid, CTO and co-founder at Finologee

How it works

MCP is an open standard from Anthropic that lets a compatible AI assistant call a platform’s tools through a common interface. Banking Orchestrator exposes around 30 tools covering its core functions. The AI assistant does not navigate the portal: it calls the tool, the Orchestrator runs the operation and the result comes back structured and ready to use.

Existing Banking Orchestrator clients can enable the MCP server through their usual Finologee contact.

Discover everything Banking Orchestrator can do at  finologee.com/banking-orchestrator.

Discover BKO

Share article on social media