Before MCP, connecting an AI assistant to an external system meant building a bespoke integration for each pairing. MCP standardizes that connection: a server advertises a set of tools (each with a name, description, and typed inputs), and any MCP-compatible client can discover and call them. The model decides which tool to call; the server runs it and returns structured results.
How Finlynq uses MCP
Finlynq ships a first-party MCP server — not a community wrapper — exposing 91 HTTP tools and 87 stdio tools across budgets, transactions, portfolios, goals, loans, subscriptions, and rules. It supports three transports:
- Streamable HTTP with OAuth 2.1 and Dynamic Client Registration — for web-based clients like Claude.ai and Claude mobile.
- HTTP with a Bearer API key — for scripts and custom agents.
- stdio — for local clients like Claude Desktop.
Because the server is first-party and open source (AGPL v3), you can audit exactly which tools exist and what each one reads or writes before you connect an assistant to your money.