Sber Unveils MCP Server for Banking AI Agents: Ouroboros Example

Olga Morozova, leader of the core components cluster at Sber API, describes how Sber developed an innovative MCP (Model Context Protocol) server for the Russian market. MCP, created by Anthropic, provides a standard interface for language models to access external tools. Sber's MCP server offers six tools: creating ruble payment orders, checking payment status, searching counterparties by TIN/BIC/name/account, and obtaining summary, account, and incremental statements. These tools can be combined to build various scenarios, such as morning financial summaries. The server uses the same security as the REST-based Sber API, with two-factor authorization via short-lived token and TLS certificate, and respects user permissions. It is model-agnostic and can work with any MCP-compatible AI agent. As an example, they connected the self-developing AI agent Ouroboros (now called GigaAgent) to demonstrate the integration. The launch aims to enable AI agents to safely handle real banking operations, opening up modular scenarios that combine banking with other functions.