APIs are the backbone of modern software. You send a request, you get a response. Simple. But there's a fundamental problem when you hand an API to an AI agent: the agent needs to know which endpoint to call, what parameters to send, and how to interpret the response.
That's where MCP — the Model Context Protocol — changes the game.
What's the difference?
Traditional API: Your code calls specific endpoints. POST /v1/refunds with a charge ID. You need to know the schema, handle authentication, parse the response, manage errors. Every integration is custom code.
MCP Server: The AI agent discovers what tools are available, understands what they do from descriptions, and calls them with natural language context. "Refund the last charge for jane@acme.com" becomes a tool call — the agent figures out which customer, which charge, and which endpoint.
Stripe's MCP vs Billy's MCP
Stripe released their own MCP server in late 2025. It's powerful — raw access to the Stripe API through MCP tools. But here's the critical difference:
Stripe's MCP is a developer tool. It gives the AI agent the same access as your secret key. No permissions. No audit trail. No spending limits. Your billing clerk can't use it — and if they could, they'd have access to everything.
Billy's MCP is a business tool. It wraps Stripe operations in permission checks, spending limits, and audit logging. When your billing clerk asks Billy to process a refund, Billy checks their role, verifies the amount is within their limit, processes it through Stripe, logs the action, and confirms. The clerk never sees an API endpoint.
Why this matters for teams
The API approach requires every team member to be technical. The MCP approach requires them to speak English (or whatever language they prefer). That's not a small difference — it's the difference between "only the CTO can touch billing" and "the support team handles billing independently."
The permission layer is everything
Without permissions, an MCP server is just a friendlier API. What makes Billy different is that the MCP tools enforce the same role-based access control as the dashboard:
- Viewer can search customers and view invoices, but can't modify anything
- Operator can process refunds within their spending limit
- Manager can manage subscriptions, create invoices, handle disputes
- Admin can do everything, including team management
Every action through MCP is logged the same way as dashboard actions. You get a complete audit trail regardless of whether someone used the dashboard, the MCP agent, or both.
The future is hybrid
We believe the best billing management experience is hybrid: a fast dashboard for bulk operations and visual overview, plus an AI agent for quick questions and one-off actions. "What's our MRR this month?" is faster to ask than to navigate to a report. "Refund order #5678" is faster to say than to find the customer, find the charge, and click through the refund flow.
MCP makes this hybrid approach possible. The AI agent and the dashboard share the same permission system, the same audit log, the same data. They're two interfaces to the same secure billing layer.
That's the future of billing management. Not choosing between a dashboard and an AI — using both, secured by the same role-based access control.