Overview

Finn is Rise Analytics’ AI analytics assistant for credit unions. It connects to your analytics data warehouse and answers questions in plain English — “How many active members do I have?” — by generating and running the right queries, then synthesizing a natural-language answer (not raw rows).

The Finn SDK (@riseanalytics/finn-sdk) is the client you use to talk to Finn from your own application. It hides the transport, auth, and streaming protocol behind a small, framework-agnostic API.

How it fits together

Your app ──▶ @riseanalytics/finn-sdk (FinnClient) ──WebSocket──▶ Finn backend ──▶ AI + data warehouse

• You call FinnClient with a URL and an API key.
• The SDK opens a WebSocket, authenticates, sends your question, and streams the answer back as it’s generated.
• The Finn backend does the heavy lifting: planning, querying the warehouse, and composing the answer.

What the SDK gives you

• Streaming. Answers arrive token by token through an async-iterable, so you can render them live.
• Conversations. Each turn returns a conversationId; pass it back to ask follow-ups with full context.
• Plain errors. Failures surface as a typed FinnError with a code you can switch on (unauthorized, rate_limited, timeout, …).
• Framework-agnostic. No framework required — it’s a plain TypeScript package (ESM + CJS) that runs anywhere Node ≥ 18 does.

Two clients, one Finn

Package	Use it when
@riseanalytics/finn-sdk	You’re writing code (a service, a script, a web backend) and want to call Finn directly.
@riseanalytics/finn-mcp	You want an AI agent (Claude, Cursor, …) to call Finn as a tool via the Model Context Protocol.

Note

The MCP server is built on top of the SDK — same connection, same streaming, exposed as an ask_finn tool. If you understand the SDK, you understand the MCP server.

Ready? Head to the Quick Start.