Why GUI May Give Way to AI Agents: Analysis Using MCP and Billing as an Example

The author, who leads billing product development at Selectel, observes that GUI design inevitably forces a choice between simplicity (handling few common scenarios) and versatility (adding filters, settings, roles), leaving users struggling when their tasks fall outside predefined paths. In billing, users typically ask intent-based questions like 'Will I have enough money until the end of the month?' but must navigate complex interfaces to find the answer. The author sees a solution in combining LLMs with agents and Model Context Protocol (MCP), an open protocol that lets AI models connect to external services and data via discoverable tools. Unlike traditional API designed for developers who know endpoints and parameters, MCP publishes capabilities declaratively for an agent to detect and invoke based on user intent. The author hypothesizes that within a few years most digital products will have an MCP server as their primary interaction layer, with GUI becoming one client among many. To test this, they built a small MCP server for Selectel billing, exposing tools for balance checks, forecasts, payments, and reports. Real-world use with the Hermes agent via Telegram showed the agent can autonomously check balances, set up proactive cron jobs, and generate expense infographics, acting closer to the user's actual intention.