# Financial Calculators ChatGPT App

A greenfield ChatGPT app/plugin starter with a TypeScript MCP server and a vanilla financial calculators widget.

The app exposes read-only tools for educational fixed-rate loan estimates. It does not use an OpenAI API key or any external financial API.

## What It Includes

- MCP server at `http://localhost:8787/mcp`
- Health endpoint at `http://localhost:8787/health`
- `calculate_mortgage` tool with structured output
- `calculate_auto_loan` tool with structured output
- Inline widget at `ui://widget/mortgage-calculator.html`
- Mortgage sliders for home price, down payment, interest rate, term, taxes, insurance, HOA, PMI, and extra payments
- Auto loan sliders for vehicle price, down payment, trade-in, interest rate, term, sales tax, fees, and extra payments
- Focused unit tests for mortgage math

## Local Development

```bash
npm install
npm run dev
```

In another terminal:

```bash
npm run inspector
```

The inspector command uses MCP Inspector against:

```text
http://localhost:8787/mcp
```

## Test In ChatGPT Developer Mode

1. Start the local server with `npm run dev`.
2. Expose it over HTTPS with Secure MCP Tunnel, ngrok, or Cloudflare Tunnel.
3. In ChatGPT, enable Developer Mode.
4. Create a developer-mode app/plugin and use the public MCP URL, ending in `/mcp`.
5. Refresh the app metadata after changing tool names, descriptions, schemas, or widget metadata.

## Example Prompt

```text
Use the mortgage calculator to estimate the monthly payment for a $400,000 home with $80,000 down, a 6.5% fixed rate, a 30-year term, $6,000 annual property tax, and $1,800 annual homeowners insurance.
```

```text
Use the auto loan calculator to estimate the monthly payment for a $42,000 vehicle with $5,000 down, a $3,000 trade-in, 7.25% APR, a 60-month term, 6.25% sales tax, and $900 in fees.
```

## Safety Note

This app is for educational estimates only. It is not financial, lending, tax, or legal advice. Production versions should add stronger disclosures, localization, logging, monitoring, and review-specific metadata before public submission.