Longevity MCP: Your Healthspan Data in Claude and ChatGPT

Your Apple Watch says your VO2 max is 41. Your Oura ring says your cardiovascular age is three years under your real age. Your Withings scale says lean mass is down a kilo since spring. Your last blood panel, a PDF in your email, says ApoB went up. Your cuff says morning blood pressure is fine.

Five apps, five numbers, five separate dashboards. Not one of them knows about the other four.

So you open ChatGPT and type "is my health getting better or worse?" and get a thoughtful, well-structured answer about general longevity principles. It cannot be about you. The model has never seen a single one of those numbers.

That gap is what this guide is about: which markers actually matter for healthspan, which ones your wearables really measure, and how to let Claude or ChatGPT read all of them at once through MCP.

Why fitness is the marker that matters most

Longevity content online tends to start with supplements and end with protocols. The research points somewhere less exciting and much more measurable: how fit you are.

Cardiorespiratory fitness predicts mortality better than most things you can track. Mandsager et al. (2018, JAMA Network Open) followed 122,007 adults who did treadmill exercise testing at the Cleveland Clinic. Lower cardiorespiratory fitness was associated with higher all-cause mortality, the association held across age groups, and the fittest group, the "elite" performers, had the lowest risk of all. There was no upper limit at which more fitness stopped being associated with lower mortality.

The pattern holds in a much larger cohort. Kokkinos et al. (2022, Journal of the American College of Cardiology) analysed roughly 750,000 US veterans aged 30 to 95. The least fit had roughly four times the mortality risk of the most fit, and the gradient held across age, sex and race.

These are observational studies. They show association, not that raising your VO2 max by a specific amount buys a specific number of years. But they are large, they agree, and they make VO2 max and the things that move it (sustained aerobic work, some high-intensity work) the obvious anchor for anyone tracking healthspan. If you want the practical side, our zone 2 training guide covers the aerobic base that most of that work sits on.

Everything else in this guide is a supporting marker. Useful, worth watching, but best read as context around fitness rather than a replacement for it.

The healthspan markers your wearables actually measure

A modern setup measures more than most people realise. The catch is that each device owns a different slice.

VO2 max and fitness age

Apple Watch estimates cardio fitness from outdoor walks and runs. Garmin estimates VO2 max separately for running and cycling and reports a fitness age. Oura estimates VO2 max and a cardiovascular age based on pulse wave velocity, a proxy for arterial stiffness.

None of these is a lab test. A wrist estimate can sit several points off a measured treadmill value. What it does well is direction: if the same device on the same kind of sessions shows a decline over six months, that trend is worth taking seriously even if the absolute number is not.

Resting heart rate and HRV

Resting heart rate falls as aerobic fitness rises and creeps up with illness, poor sleep, alcohol and accumulated stress. HRV moves in the opposite direction for most of the same reasons. Single readings are noisy. Seven-day averages against a 30 to 60 day baseline are where the signal lives. Our guides on resting heart rate trends and HRV-guided training go deeper on reading both.

Sleep timing

Most people look at sleep duration and a score. For long-term health, the steadiness of when you sleep is easier to act on than any stage breakdown. A bedtime that swings by two hours between weekdays and weekends is something you can see in a month of data and change next week. Consumer stage estimates are the least reliable part of wearable sleep data, so lean on the parts the device measures well. The sleep and training guide covers what wearables get right and wrong.

Body composition

Lean mass tends to decline with age, and keeping it is one of the clearer healthspan goals. Smart scales from Withings and Renpho report fat mass, muscle mass and, on some models, segmental readings. Bioimpedance is imprecise on any single morning and influenced by hydration, so weekly averages and multi-month direction are the honest way to read it. A DEXA scan, if you have one, is a much better anchor point.

Blood pressure and ECG

Blood pressure is one of the best-established cardiovascular risk factors, and a home cuff gives you far more readings than an annual check-up. Withings ScanWatch records single-lead ECGs and classifies each one as normal, AFib or inconclusive. Neither replaces clinical assessment, but a history of readings is exactly what your doctor will want to see.

Bloodwork

Lipids, ApoB, HbA1c, ferritin, vitamin D, thyroid markers. No wearable measures these. They arrive as a PDF once or twice a year, and the useful question is almost never "is this panel normal?" It is "how does this panel compare with the last three, and did anything I changed in between line up with the move?"

Why one dashboard is never enough

Every longevity app solves this by picking a subset of markers and scoring them in its own interface. WHOOP Healthspan lives inside WHOOP, and it needs a Peak or Life membership. Oura shows Cardiovascular Age inside Oura. Garmin shows fitness age inside Garmin Connect. Dedicated longevity apps such as Vora or Wellness Project pull data from several devices and score it their way.

Those views are fine. The limit is that each one only sees what it ingests, and each one decides for you which questions are worth asking.

A general model like Claude or ChatGPT works the other way round. You ask the question, in your words, and it reasons over whatever it can read. "Did my resting heart rate and blood pressure both improve after I added a third aerobic session in March?" crosses a wearable, a cuff and your training log. No single app holds all three.

The model is only as good as its access, though. That is the whole problem, and it is what MCP solves.

What MCP changes

The Model Context Protocol is an open standard that lets AI assistants call external tools. A server advertises tools, the model decides when to call one, structured data comes back. Claude, ChatGPT, Cursor and Perplexity all support it.

For health data, that means the model fetches the window it needs when it needs it. Ask about VO2 max over a year and it calls a VO2 max tool for that year. Ask whether your last two blood panels differ and it lists your uploaded files and reads both. Nothing gets pasted, nothing goes stale, and the answer is computed from your numbers rather than from general knowledge. Our MCP overview explains the protocol and the setup in more detail, and the best fitness MCP servers guide compares every hosted and self-run option.

Your options, honestly compared

There are now several ways to get health data into Claude or ChatGPT. They differ a lot in what they can read and where they work.

ChatGPT Health. Built into ChatGPT, launched in 2026. It can read Apple Health from an iPhone and connect medical records. It is US-only and 18+, and as of writing it is not available in the EU, UK or Switzerland. If you live in the US and your data is all in Apple Health, it is the lowest-effort route.

Claude with Apple Health on iPhone. A beta for Pro and Max plans in the US, read-only. Same geography limit, same dependency on Apple Health as the only source.

Fulcra. A personal data store fed by its iPhone app: Apple Health plus location and calendar, with a hosted MCP server and an open-source local one. It does not connect wearable APIs directly, so anything that does not flow into Apple Health is out of reach.

freddy. A hosted, read-only MCP server listed in Claude's directory and available as a ChatGPT app. Many of its sources arrive through Apple Health or Health Connect. It is a data pipe with no coaching layer.

Longevity apps. Vora, Wellness Project, WHOOP Healthspan and Oura Cardiovascular Age score your data inside their own product. Wellness Project also offers an MCP server. Good if you want a ready-made score, limited if you want to ask your own questions across every device.

athletedata. A hosted MCP server with one personal URL, also listed in Anthropic's Claude Connectors Directory, that works in Claude, ChatGPT (custom MCP connectors need a paid ChatGPT plan), Cursor, Perplexity and Codex, in any country. It reads 39 apps, from direct wearable connections to the Apple Health and Health Connect phone apps, and it adds a Health Vault for documents. The longevity-relevant tools include:

• oura_get_vo2_max and oura_get_cardiovascular_age for Oura's fitness and arterial-stiffness estimates
• apple_health_get_vo2max, apple_health_get_resting_hr and apple_health_get_hrv for Apple Watch history
• garmin_get_user_metrics for Garmin VO2 max (running and cycling) and fitness age
• withings_get_heart_list for ScanWatch ECG recordings and their classifications, and withings_get_measurements for weight, fat mass and muscle mass
• renpho_get_body_composition for Renpho weigh-ins including lean mass
• get_blood_pressure for cuff readings from Withings, Garmin Index BPM, or any cuff that writes to Apple Health or Health Connect
• eight_sleep_get_sleep for Eight Sleep Pod nights, and get_sleep_trend for sleep aggregated by week or month and compared before and after a change
• list_athlete_files and read_athlete_file for lab panels, DEXA scans and other documents you upload

What athletedata does not have. No continuous glucose monitor data and no DNA or genetic results. If glucose response is central to your longevity work, you need a different tool for that part. Strava is also not served over our MCP because Strava's API terms prohibit it; Strava has its own official connector for subscribers.

How bloodwork fits in

Lab results are the piece most people never get into a model properly. Pasting one panel into ChatGPT works for a single read. It falls apart when you want three years of panels compared against your training.

The Health Vault takes PDFs, images and CSVs. Each upload is read once and stored as an extracted summary of the key values and the date the sample was taken. Over MCP, list_athlete_files shows the model what you have and read_athlete_file returns those summaries, so "compare my last two blood panels" becomes two tool calls and a side-by-side of the markers the summaries carry. A summary is not the whole file: a dense panel with dozens of rows can lose some of them, so treat an absent marker as "not in the summary", not "not on the panel". CSV uploads are the exception, because the model can read their exact rows.

Two honest caveats. Extraction from a scanned PDF can misread a value, so check any number that matters against the original before you act on it. And the model's read of your lipids or HbA1c is a starting point for a conversation with your doctor. It is not an interpretation of your results, and it should never be the basis for starting, stopping or changing a medication.

Where this gets genuinely useful is the cross-reference. "My ferritin dropped between the April and October panels. Did my training volume or resting heart rate change over the same months?" That question needs a lab PDF and a wearable in the same conversation. Very few setups can answer it.

Example prompts that work

Once the connector is added, ask in plain language. These are the questions people actually use:

• "Plot my VO2 max and resting heart rate over the last year and tell me what moved it."
• "Compare the markers you can read from my last two blood panels, list the ones that changed by more than 10% with their dates, and tell me which values you could not find in the summary so I can check the PDF."
• "My Oura cardiovascular age jumped up in September. What else changed that month: sleep timing, HRV, training load, alcohol tags?"
• "Show my blood pressure readings for the last 90 days as a range per week, not an average, and flag any week where the morning readings shifted."
• "Has my lean mass changed since I started lifting twice a week in June? Use weekly averages from the Withings scale."
• "How regular is my bedtime? Give me the spread between my earliest and latest sleep onset per week for the last two months."
• "Summarise my last three ECG recordings and their classifications so I can bring them to my cardiologist."
• "Draft five questions for my doctor based on my last two blood panels and the resting heart rate trend since then."

The last one is the best use of the whole setup. A model that has read a year of your data writes far better questions than you will at 8am in a waiting room.

WHOOP vs Oura for longevity tracking

If you are choosing a device with longevity in mind, the honest answer is that consistency beats brand. A ring or strap you wear every night for three years produces a usable trend; the better sensor sitting in a drawer does not.

On data access, there is a real difference. Oura's API exposes VO2 max and cardiovascular age alongside sleep timing, HRV and temperature, so a model can read them directly. WHOOP's Healthspan view sits inside the WHOOP app behind a membership tier, while recovery, HRV and sleep are available to outside tools. If you want an AI to reason over your longevity estimates rather than just your recovery, that tilts toward Oura. Our Oura vs WHOOP comparison covers the rest, and the Oura MCP server guide lists every Oura tool.

Setup and pricing

1. Create an athletedata account and connect the devices you use on the integrations page. Apple Health and Samsung Health need the athletedata phone app to sync.
2. Upload any lab PDFs, DEXA scans or old test reports to the Health Vault.
3. Copy your personal MCP URL from the dashboard and add it as a connector in Claude or ChatGPT. Step-by-step instructions are on the Claude setup and ChatGPT setup pages.
4. Ask a first question that needs history, such as the VO2 max prompt above, and check the answer against what you see in the device app.

The MCP tier is $9 per month or $69 per year, with a 7-day free trial, and every tool in this guide is included. If you want training built around the trends, the coaching tier at $39 per month adds a proactive coach on WhatsApp, Telegram or iMessage that adapts your plan to your recovery and fitness data. Details are on the pricing page.

Putting it together: a healthspan data protocol

1. Pick one VO2 max source and stick with it. Compare the same device on the same kind of sessions, and judge it by six-month direction, not single readings.
2. Track resting heart rate and HRV as 7-day averages against a 30 to 60 day baseline. Ask the model to flag sustained shifts, not daily noise.
3. Watch bedtime and wake-time spread for a month. If it swings by more than an hour, fix that before chasing any other sleep metric.
4. Weigh in at the same time of day and read lean mass as a weekly average over months. Anchor it with a DEXA scan if you can.
5. Take blood pressure readings at consistent times and read the range per week, never one daily average.
6. Upload every blood panel to one place, then ask the model to compare each new panel with the previous ones and with your wearable trends over the same period.
7. Once a quarter, ask the model to draft questions for your doctor from all of it, and take that list to your next appointment.
8. Keep the AI on its side of the line: trends, summaries and questions. Diagnosis and medication stay with your doctor.

Most of what predicts a long, healthy life is already being measured on your wrist, your finger and your bathroom floor. The work is getting it into one place where someone, or something, can actually read it.