Ask Claude or ChatGPT to Fill In Your Retirement Calculator

A reader emailed me last month. She had a UK State Pension, a small teacher's defined benefit pension from Scotland, and a Fidelity IRA from eight years she'd spent working in Massachusetts before coming home. She'd been chatting with ChatGPT about it for months. She knew her numbers. She'd asked the AI to estimate when she could retire, what her monthly income would look like, whether the Fidelity account would outlast her.

Then she found RetireFlexi and spent three evenings entering everything again by hand.

She did not need to. The AI already had everything. She just needed to ask it to hand the information across.

You may have read about "MCP servers" — ignore that

If you've been following AI news, you've probably come across something called an MCP server. Articles describe it as a way to connect AI assistants directly to your applications. Every article I've read on the subject seems written for software developers who already know what a port number is.

It isn't what you want. An MCP server is a technical integration layer that requires installing packages on your computer, running a local server process, configuring authentication. If none of those words mean anything to you, good — you don't need to touch any of it. That's a developer's tool for building AI products, not a solution for someone who wants their pension figures in a retirement calculator.

The approach that actually works for a non-developer is considerably simpler.

How this actually works

RetireFlexi was built for this from the start.

At retireflexi.com/llms-full.txt there's a plain-text file that explains the calculator — every field, every section, what each one means, what format plans are stored in. When you ask Claude or ChatGPT to read that URL, they do. They come back knowing exactly what RetireFlexi expects, without any configuration on your end. AI assistants can fetch webpages; this is a webpage. That's the mechanism.

Plans are stored as plain-text files — readable, not proprietary, not encrypted. The AI generates one. You load it into the calculator via the Open Existing Plan button. That's the whole process.

Step by step: how to do it

Step 1: Open your existing AI conversation

Use Claude at claude.ai or ChatGPT at chatgpt.com. Either works. Open the chat where you've been discussing your retirement — or start a new one and describe your situation now.

The more specific you've been, the better the output. If you've told it your pension amounts, the countries involved, and when you want to retire, it has most of what it needs.

Step 2: Ask the AI to read RetireFlexi's instructions

Copy and paste this into your chat:

Please read the full documentation at https://retireflexi.com/llms-full.txt so you understand exactly what RetireFlexi expects. Then, based on everything we've discussed about my retirement, generate a RetireFlexi plan file in the correct JSON format. Include all the pension sources, savings accounts, and income streams we've talked about. Use your best estimates for any fields I haven't specified — I'll review and correct them once the plan is loaded. Save the result as a file I can download.

That's the whole instruction. The AI will fetch the documentation, understand the format, and use your conversation history to populate the fields.

Step 3: Download the file the AI gives you

The AI will either offer a download link or display the plan as a block of text in the chat. If it's text rather than a download, ask it: "Give me this as a downloadable file." It will. This is occasionally fiddly in ChatGPT — if the download button doesn't appear the first time, just ask again.

Save the file somewhere easy to find. Your Desktop is fine.

Step 4: Load it into RetireFlexi

Go to retireflexi.com. On the opening screen, look for the Open Existing Plan button (it's in the top toolbar, or on the welcome screen if you're starting fresh). Click it, find the file you just saved, and open it.

Your retirement figures will load immediately. You'll see your pensions listed, your accounts populated, and the dashboard will run your projections automatically.

Step 5: Check the numbers

The AI will get the structure right — the right sections, the right account types, pensions in the right places. It will get some of the specific numbers slightly wrong, because it was working from your conversation rather than your actual statements.

Go through each section and verify against your real figures. Growth rate assumptions are the most likely to need adjustment — AI assistants tend to use conservative defaults that may not match your actual fund performance. Pension ages and escalation rates are also worth checking.

This verification takes twenty minutes. Building the plan from scratch takes three evenings. Even with corrections, you're well ahead.

What the AI actually produces

If you're curious what you'll be looking at, a RetireFlexi plan is a JSON file — which sounds technical but is just a structured list of your financial details. It looks something like this (simplified):

{
  "setup": {
    "name": "My retirement plan",
    "currentAge": 62,
    "retirementAge": 67,
    "horizonAge": 90,
    "drawdownCurrency": "GBP",
    "taxCountry": "United Kingdom",
    "targetAnnualIncome": 35000
  },
  "statePensions": [
    {
      "country": "United Kingdom",
      "annualAmount": 11502,
      "drawAge": 67,
      "annualEscalation": 3
    }
  ],
  "retirementSpecific": [
    {
      "country": "United States",
      "currency": "USD",
      "productType": "IRA",
      "currentValue": 145000,
      "annualContributions": 0,
      "expectedGrowth": 6
    }
  ]
}

The AI generates the complete version of this — every section filled in based on what you told it. You don't need to understand the format. You just load the file and RetireFlexi reads it.

Your data doesn't go anywhere it hasn't already been

When you ask Claude or ChatGPT to generate this plan, your financial details are in the conversation with that AI service — which you've already been using. That's unchanged. The file it generates sits on your computer. When you load it into RetireFlexi, everything runs in your browser: RetireFlexi's server never sees your figures, never stores them, and has no database of retirement plans.

The only thing RetireFlexi's server does when you open the calculator is deliver the calculator itself to your browser. After that, your data goes nowhere.

If you then choose to save your plan by emailing it to yourself, RetireFlexi handles the email delivery — but the server processes the email without reading or retaining the attachment. More detail on how this works is on the about page.

What the AI handles well — and where it guesses

Getting the structure right is easy for an AI. It knows the difference between a State Pension and a defined benefit scheme, between an IRA and a 401(k), between a SIPP and a UK State Pension. If you've told it about these things in your conversation, it will put them in the correct sections.

Where it guesses is in the assumptions. If you've told the AI you have a Fidelity IRA but haven't specified the expected growth rate, it will pick something reasonable — probably 5–7% nominal. That may or may not match how your actual funds have performed or how you want to model future performance. Same for inflation rate, care cost assumptions, and the go-go/slow-go spending phases that RetireFlexi models through retirement.

The expat retirement planning guide covers what these assumptions mean and what figures are reasonable for different account types. That's worth reading before you finalise your plan's numbers.

One thing the AI definitely cannot know: the current value of accounts it has no information about. If you never told it your SIPP balance, it will put in a placeholder or ask you. That's fine. Load the plan, update the figure, done.

If you haven't started chatting with an AI yet

Everything above assumes you've already had some retirement conversation with an AI assistant. If you haven't, you can still use this approach — you'll just be doing the describing and the plan generation in the same conversation.

Start with something like this:

I want to model my retirement in a calculator called RetireFlexi. Please read https://retireflexi.com/llms-full.txt first so you understand the format. Then ask me questions about my retirement situation — my age, pension sources, savings accounts, target income, and where I plan to retire. Once you have enough information, generate a RetireFlexi plan file I can download and load into the calculator.

The AI will walk you through the questions in a conversational way, which many people find easier than looking at a form cold. Once you've answered, it generates the plan and you load it.

For plans involving multiple countries

RetireFlexi handles UK, US, Australian, Canadian, and dozens of other pension systems. If your retirement picture spans two countries — a UK State Pension and a US 401(k), for example — make sure your AI conversation has covered both clearly before asking for the plan file.

The US Social Security and UK State Pension guide explains how these two systems interact, which affects how you should enter them into RetireFlexi. Specifically, US Social Security and UK State Pension are entered as separate state pension entries, each with their own draw age and escalation rate. The AI will do this correctly if you've told it you have both — but it's worth verifying.

If you're considering retiring to another country, the cost of living comparison guide gives the current Numbeo figures for the most popular expat retirement destinations. Japan's costs dropped 46% since 2021. Mexico got 21% more expensive. These are the numbers to have in mind before you ask the AI to model your budget for a specific destination.

The output reflects the conversation

If you told the AI your SIPP balance was £180,000 back in November, it'll use £180,000. If you mentioned you have 28 of 35 qualifying years for the State Pension, it'll put that in. If you were vague — "I have some savings and a pension from work" — the plan will be vague too.

That's not a reason to wait until you've had the perfect conversation. Load what the AI generates, go through each section against your actual statements, and fix what's wrong. The structural work — getting the right account types in the right sections, separating a DB pension from an IRA from a State Pension — is the part that takes time to do manually. That part it gets right. The specific numbers you check in twenty minutes. Three evenings versus twenty minutes. That's still the comparison even on a rough first draft.