Loanie is AI-ready

Connect Loanie to yourMCP-compatible agent

Loanie publishes a public `mcp.json` manifest and a remote MCP server, so your assistant can work with NZ mortgage rates, property data, repayment maths, break-fee estimates, referral codes, and loan applications from one connection.

Also published for agents:`skills.md``function-calling`

Manifest

`/.well-known/mcp.json`

Use the well-known manifest when your client supports remote MCP import.

Transport

streamable-http

Connect directly to https://loanie.co.nz/api/mcp when your client wants the raw server URL.

Coverage

Rates, property, applications

The published toolset covers discovery, calculations, referrals, and application submission.

Tool coverage

What your AI can do with Loanie

The manifest at `public/.well-known/mcp.json` advertises a mortgage workflow that is already structured for agent use, not just content retrieval.

Live mortgage rates

Compare current NZ mortgage rates across major banks without scraping lender websites by hand.

Property search and details

Search addresses, resolve property IDs, and fetch valuation ranges, bedrooms, bathrooms, land area, and images.

Repayment modelling

Calculate weekly, fortnightly, and monthly repayments from principal, rate, and term.

Borrowing capacity

Estimate how much a borrower may be able to borrow using income, expenses, deposit, and a stress-test rate.

Break fee estimates

Estimate early repayment costs for fixed-rate mortgages with NZ bank-specific inputs.

Applications and referrals

Look up invitation codes and submit loan applications directly to Loanie with only the minimum required details.

How to connect

1. Start with the manifest

Use the well-known manifest URL if your MCP client supports remote manifest import. That file tells your agent where Loanie lives and what it can do.

https://loanie.co.nz/.well-known/mcp.json

2. Fall back to the direct server URL

If your tool asks for a raw MCP endpoint instead of a manifest, connect directly to the Loanie server over streamable HTTP.

https://loanie.co.nz/api/mcp | transport: streamable-http

3. Ask for mortgage work in plain English

Once connected, your assistant can compare rates, search properties, estimate repayments, calculate break fees, resolve invitation codes, and submit applications.

Example: "Compare 1-year special rates from ANZ, ASB, BNZ, Westpac, and Kiwibank."

4. Parse tool output as JSON text

Loanie returns tool results as text that contains a JSON string. If you are building your own client, parse that text field before using the data.

response_conventions.format = JSON.stringify inside content[0].text

Manifest preview

Minimal config, real workflow

If your MCP client shows raw configuration instead of a UI import flow, these are the important fields.

{
  "mcpServers": {
    "loanie": {
      "url": "https://loanie.co.nz/api/mcp",
      "displayName": "Loanie - NZ Mortgage & Loan Platform",
      "transport": "streamable-http",
      "capabilities": { "tools": true },
      "docs": {
        "skills": "https://loanie.co.nz/skills.md",
        "functionCalling": "https://loanie.co.nz/.well-known/openai-function-calling.json"
      }
    }
  }
}

Developer note

Loanie's MCP responses are returned as text containing a JSON string. Parse `content[0].text` before you use the result in your own app or orchestration layer.

OpenAI function calling

Fetch a ready-to-use tools array

The published JSON follows the OpenAI function-calling guide, enables strict mode on every definition, and converts optional inputs into strict-compatible nullable fields.

import OpenAI from "openai";

const openai = new OpenAI({ apiKey: process.env.OPENAI_API_KEY });
const { tools } = await fetch("https://loanie.co.nz/.well-known/openai-function-calling.json").then((res) => res.json());

const response = await openai.responses.create({
  model: "gpt-5.4",
  input: "Compare current 1-year special NZ mortgage rates.",
  tools,
  tool_choice: "auto",
  parallel_tool_calls: false,
});

skills.md

Give agents workflow guidance before they touch tools

skills.md adds task-level instructions on when to research first, when to confirm user intent, and which actions create a real Loanie application.

# Loanie Agent Skills

## Preferred discovery endpoints
- MCP: https://loanie.co.nz/.well-known/mcp.json
- OpenAI: https://loanie.co.nz/.well-known/openai-function-calling.json

## Skills
- Compare NZ mortgage rates
- Search properties and fetch details
- Calculate repayments and borrowing capacity
- Estimate break fees
- Check invitation codes
- Submit loan applications with confirmation

All three assets describe the same Loanie workflows: rates, property search, calculators, referral checks, and loan-application handoff.

Prompt ideas

Start with these requests

Compare current 1-year special mortgage rates from the major NZ banks and rank them lowest to highest.
Search for 12 Example Street, Auckland and show the best property match with estimated value and bedrooms.
Estimate repayments for a NZD 780000 mortgage at 5.89 percent over 30 years.
Estimate the break fee for a Westpac fixed loan with my current rate, balance, and fixed-term start date.
Check invitation code ABC123 and tell me who sent it and whether it already has a property attached.
Start a refinance application for Jane Smith with email [email protected] and a property value of NZD 950000.

Guardrails

Built-in safety expectations

Do not request or transmit internet banking credentials.

Collect only the minimum personal details needed for loan applications: full name, email, phone optional, property address, property value, deposit, and referral code.

Avoid government ID numbers, CVVs, or bank passwords. Those are explicitly disallowed in the published policy.

Best practice

Use AI to compress admin and surface options quickly, then hand the customer off to the appropriate Loanie workflow for actual application handling and advisor follow-up.