MCP server

Connect AI agents to FinologeeBKO through a Model Context Protocol endpoint that shares the REST API's authentication, rights, and approval model

Introduction

The MCP server exposes FinologeeBKO to AI agents, such as Claude, ChatGPT, or Mistral, through the Model Context Protocol (MCP). Agents connected to this endpoint use the platform in the same way an integration built on the Payments API would, but through natural-language interaction rather than direct API calls.

The MCP endpoint does not introduce a separate permission model. Every action an agent takes runs under the authenticated user's existing rights, and the agent can only successfully execute what that user is permitted to do in FinologeeBKO. Payment and other data-changing actions follow the same signatory-rule approvals that apply to the equivalent action in the user interface or the REST API.

This approach lets treasury teams, fund administrators, and other users delegate repetitive or exploratory tasks, such as listing unauthorized payments or drafting a payment based on an invoice, to an AI agent, while keeping the platform's existing controls in place.



Prerequisites

The MCP server is available to users who already have access to FinologeeBKO through the standard authentication and rights model. No separate MCP-specific permission exists; access to a given tool follows the same permissions used elsewhere on the platform.

Feature flags

Feature FlagDescription
MCPEnables the Model Context Protocol endpoint for the tenant

The MCP server is enabled per tenant. Contact [email protected] to request it for your organization.



MCP server overview

The MCP endpoint provides approximately 40 tools covering the following functional areas:

  • Payments (credit transfers)
  • Direct debits
  • Cash and accounts
  • Counterparties
  • Invoicing
📘

Scope of the MCP server

The first release of the MCP server does not include securities or cash-concentration tools. Coverage for these areas may be added in a later release.

Rights and approvals

Every tool invocation runs under the identity and rights of the connected user. An agent can only successfully complete an action that the user is permitted to perform in FinologeeBKO. Where an action would normally require signatory-rule approval, such as authorizing a payment, the same approval requirement applies when the action is triggered by an agent.

Audit trail

Data-changing actions performed through the MCP server are recorded in the audit trail in the same way as the equivalent action performed through the REST API.



Connecting an AI agent

An agent connects to the MCP server as a custom connector or MCP client. Before you start, you need two values that identify your tenant and authenticate you:

  • MCP endpoint URL — your FinologeeBKO host with /mcp appended, for example https://<your-tenant-host>/mcp
  • OAuth client ID — the identifier for your tenant's connection. You can find it in the client_id query parameter of your FinologeeBKO login page URL, or request it from [email protected]

Authentication uses the OAuth authorization-code flow: when you connect, a browser opens at the FinologeeBKO login page, and you sign in with your usual credentials. The connection uses a public client, so no client secret is required. Once connected, the agent's tools appear namespaced as bko__<domain>__<verb>, for example bko__cash_account__find_cash_accounts.

Select your AI agent for step-by-step setup instructions:



Best practices

  • Review agent-initiated actions - Treat data-changing actions taken by an agent with the same scrutiny as actions taken through the user interface or API
  • Grant rights deliberately - Since the MCP server enforces the connected user's existing rights, review user permissions before connecting an AI agent on their behalf
  • Rely on existing signatory rules - Do not treat agent-initiated payments as exempt from approval; signatory-rule approvals apply regardless of how a payment was created


Related documentation

Explore related sections for more information:



Support

For assistance enabling the MCP server for your tenant or questions about AI agent integration, contact [email protected].


Did this page help you?