# Getting started

> From sign-up to an agent reading your money: connect accounts, add context, connect Claude or ChatGPT, read the receipts.

Canonical: https://usetruenote.com/docs/getting-started

## 1. Make an account

Go to [app.usetruenote.com](https://app.usetruenote.com/sign-in) and type your email. A six-digit code signs you in and makes the account; the first 14 days are free. Save a passkey in Settings afterwards and you won't need codes again.

## 2. Connect your accounts

Accounts → **Connect an account** opens Plaid Link. Sign in at your bank there (Truenote never sees the login). Accounts appear within seconds; history imports over the next minute or so (Plaid sends about 30 days first, then the rest). A house, a car or anything Plaid can't reach goes in as a manual account with dated balances.

Accounts you never use are hidden after a sync so they stay out of totals; Settings → Hidden accounts shows them again.

## 3. Add the context

The welcome steps ask for the basics: who's in the household, each income source, one goal. Then:

- **Plan**: regular take-home and how it's split into groups and lines.
- **Buckets**: a virtual split of a savings account, each with a ledger and an optional target. Deposits are split by what each bucket is due; spending is charged by rules you approve in **Review**.
- **Context**: businesses (so their money is reported apart), categories and rules, notes, plans.
- **Taxes**: both W-2s, self-employed income, bonuses and estimated payments → the year's projection. Upload a paystub under **Documents** and its figures are offered as the inputs.

All of it can be written by an agent instead, once one is connected.

## 4. Connect an agent

Connected agents shows the MCP URL and the steps for each client:

- [Claude](/connect/claude): Settings → Connectors → Add custom connector.
- [ChatGPT](/connect/chatgpt): Settings → Apps → developer mode → Create app.
- [Claude Code](/connect/claude-code): `claude mcp add --transport http truenote https://app.usetruenote.com/mcp`.
- [Scripts](/connect/api): a personal access token and `/v1`.

A consent screen shows what the agent may see; you can hide accounts or a business from it. New connections start read-only. The first time an agent wants to write, it asks and you approve again.

## 5. Read the receipts

Connected agents → **Activity** lists every call: the agent, the tool, a summary of the arguments, the row count, refusals. **Revoke** cuts an agent off on its next call.

## Things to ask

See [Prompts that work well](/docs/prompts).