hi folks,

I've posted a few data breakdowns here over the years. This one is less of a breakdown and more of a setup: I've got a race on Sunday, and this is what's been deciding my sessions in the run-up.

Last week I didn't do a single one of the workouts I had planned, which sounds worse than it is. Tuesday was 34 degrees so I swam instead, and Saturday's long run was meant to be an hour and a half but I only had an hour, so I ran it harder.

Both were the right call at the time. The problem is that neither of them made it back into the plan, and if you keep doing that for a month you've properly lost track of what you've actually been training.

So there's a check that runs every morning now, and it goes off what I actually did. It picks up the swim sitting where a run should have been, notices the long run came in short and fast, and works today's session out from there.

It's a Claude project holding my plan, with a scheduled task each morning. It pulls yesterday's workouts with type, duration and Strain, this morning's Recovery including HRV and resting heart rate, and last night's sleep against what WHOOP reckoned I needed. Then it looks at today's planned session and commits to one call: it holds, or it changes and the reason goes in the calendar event.

The data comes through freddy, an MCP server I built that connects your health and performance data to whatever AI you already use. Openly my project.

Write-up with the prompts: https://freddy.coach/recipes/adaptive-triathlon-plan

Two things that bit me. Recovery lands in the morning but your workouts land immediately, so tell it to say when it has yesterday's session without yesterday's sleep numbers rather than filling the gap. And never let it compare HRV across sources, WHOOP reports RMSSD while Apple Health reports SDNN, so a chart mixing them looks meaningful while being nonsense. It isn't a coach or a clinician either.

WHOOP, Polar, Garmin, Oura, Suunto, Withings, Wahoo, Ultrahuman and Intervals.icu are supported, with iPhone and Android apps for Apple Health or Health Connect.

I know health data is sensitive and I have handled it for years with fitIQ. Data is encrypted, I do not sell it, and I am not looking to make a profit off your stats, but if you still don't trust 3rd party solutions, just don't use it :)

Hi folks,

I posted here a while back when Suunto support landed in freddy, the personal health MCP. Coming back with something more useful than an integration announcement: I've got a race on Sunday and this is what's kept the plan honest.

Last week I didn't do a single one of the workouts I had planned, which sounds worse than it is. Tuesday was 34 degrees so I swam instead, and Saturday's long run was meant to be an hour and a half but I only had an hour, so I ran it harder.

Both were the right call at the time. The problem is that neither of them made it back into the plan, and if you keep doing that for a month you've properly lost track of what you've actually been training.

So there's a check that runs every morning now, and it goes off what I actually did. It picks up the swim sitting where a run should have been, notices the long run came in short and fast, and works today's session out from there.

https://preview.redd.it/8zl1oz1qn4kh1.png?width=3200&format=png&auto=webp&s=7edff34fc86bf69b60df2425672cdf35cbbf975a

It's a Claude project holding my plan, with a scheduled task each morning. It pulls yesterday's workouts with type and duration, the Training Effect and recovery time Suunto put on them, last night's sleep and HRV, and how my stress sat through the day. Then it looks at today's planned session and commits to one call: it holds, or it changes and the reason goes in the calendar event.

The data all comes through freddy, which connects your sources to whatever AI you already use. Openly my project.

Write-up with the prompts: https://freddy.coach/recipes/adaptive-triathlon-plan

Worth knowing: run it after your watch has synced, because last night's sleep doesn't exist until the watch uploads it, so an early run reports on the night before last. Tell it a single bad night is noise, or it turns cautious after one poor sleep. And it isn't a coach or a clinician.

Suunto, Polar, Garmin, Oura, WHOOP, Withings, Wahoo, Ultrahuman and Intervals.icu are supported, with iPhone and Android apps for Apple Health or Health Connect.

I know health data is sensitive and I have handled it for years with fitIQ. Data is encrypted, I do not sell it, and I am not looking to make a profit off your stats, but if you still don't trust 3rd party solutions, just don't use it :)

I've posted about freddy here before, the personal health MCP. This time I wanted to share how I've actually been using it myself, prepping for a race this Sunday.

Last week I didn't do a single one of the workouts I had planned, which sounds worse than it is. Tuesday was 34 degrees so I swam instead, and Saturday's long run was meant to be an hour and a half but I only had an hour, so I ran it harder.

Both were the right call at the time. The problem is that neither of them made it back into the plan, and if you keep doing that for a month you've properly lost track of what you've actually been training.

https://preview.redd.it/9l418vfni3kh1.png?width=3200&format=png&auto=webp&s=f7070d38f6bec41b845b23f497fbf355aee6bd4e

So there's a check that runs every morning now, and it goes off what I actually did. It picks up the swim sitting where a run should have been, notices the long run came in short and fast, and works today's session out from there.

It's a Claude project holding my plan, with a scheduled task each morning. It pulls yesterday's workouts with type, duration and intensity, last night's Nightly Recharge including ANS Charge and HRV, and my Cardio Load over the last week against my own 30 day baseline. Then it looks at today's planned session and commits to one call: it holds, or it changes and the reason goes in the calendar event.

The data all comes through freddy, which connects your sources to whatever AI you already use. Openly my project.

One Polar note: a plan is only worth building if it's justified against your real training, and the old AccessLink API only hands over about 30 days. freddy connects Polar's newer API too now, so years of past sessions and Nightly Recharge come through.

Write-up with the prompts: https://freddy.coach/recipes/adaptive-triathlon-plan

Two things worth knowing. Make it commit to one call, because "listen to your body" is what you already had. And never let it compare HRV across sources, RMSSD and SDNN are different measurements rather than different units, so mixing them looks meaningful while being nonsense. It isn't a coach or a clinician either.

I know health data is sensitive and I have handled it for years with fitIQ. Data is encrypted, I do not sell it, and I am not looking to make a profit off your stats, but if you still don't trust 3rd party solutions, just don't use it :)

Fun one to set up over a weekend if you run OpenClaw agents. I build freddy, a personal health data server that exposes your health and performance data as an MCP server, and it just went live as an app on iOS and Android.

The idea

Because it is plain MCP, an OpenClaw agent can connect to it and reason over your actual body data instead of generic advice. You connect your sources once, and the agent can pull sleep, recovery, training load, and the rest inside whatever workflow you are running.

https://preview.redd.it/264uymhe9kch1.png?width=2540&format=png&auto=webp&s=8c6dcf0302ba44abba6e344a1583163e849f8621

What you can connect

Apple Health or Health Connect from the phone, plus Oura, Polar, WHOOP, Garmin, Withings, Suunto, Intervals.icu, and Hevy.

Getting an agent in

For a headless agent the device flow is the clean way in, so the agent gets its own scoped access without you pasting a URL around.

A weekend project idea

A morning agent that looks at last night's sleep and your recent training load and writes you a two line note on whether to push or back off today. It is a small thing to build on top and it reads real numbers rather than guessing.

Link

You can find it at https://freddy.coach/.

On privacy

I know health data is sensitive and I have handled it for years with fitIQ. Data is encrypted, I do not sell it, and I am not looking to make a profit off your stats, but if you still don't trust 3rd party solutions, just don't use it :)

I have worn a WHOOP for years and it is the best thing I own for recovery, strain, and training readiness. Two years ago I posted about my transformation here and it's partially (a big part) thanks to WHOOP.

What I was missing is the context of my training performance data (runs, swims, rides) or my environment data (temperature, CO2 levels etc). So I built freddy (freddy.coach), a personal health data server, and recently opened it up to everyone.

freddy is an MCP server. It connects your health data to any AI that supports MCP: Claude, ChatGPT, Gemini, Perplexity and others. You paste a personal URL into your AI client once, and after that it can read your data in any conversation.

https://preview.redd.it/x7jo1g8dczbh1.png?width=2400&format=png&auto=webp&s=4f7202be1d37a9580da887a4a03c07718558271a

The news for this sub: WHOOP is now out of beta, and open to everyone in freddy. Once you connect it, it pulls your recovery, strain, sleep, HRV, resting heart rate and workouts.

Oura, Garm in, Polar, Withings, Suunto, Intervals.icu, Hevy, Apple Health (iOS), and Health Connect (Android) are supported too, most of them open to everyone now. You connect whatever you actually use.

There is an always free tier, so no risk to check it out at freddy.coach

Yes, you can connect WHOOP through API to Claude or other agents without intermediary, or use any of the other 100 recently vibe-coded "dashboards" or MCPs. If getting WHOOP data alone is your goal, freddy is not for you. It only makes sense if you have other sources you want to bring in and analyze together.

I know health data is sensitive and I have handled it for years with fitIQ. Data is encrypted, I do not sell it, and I am not looking to make a profit off your stats, but if you still don't trust 3rd party solutions, just don't use it :)

Hey folks, some of you know freddy already, a personal health MCP I built to solve my own problem. freddy actually launched here, and the reception has been amazing, so I keep working on it. Just added Airthings environment data integration and full CLI support with an npm package.

freddy connects your health and performance data to an AI like Claude, ChatGPT, Gemini or Perplexity. You paste one personal URL into your AI client, and from then on it can read your data in any conversation. Polar has been supported since the start, along with Oura, Suunto, Withings, Intervals.icu, Concept2, Wahoo and Hevy, and more.

https://preview.redd.it/v13p9s2quz4h1.png?width=1144&format=png&auto=webp&s=f60fb8ea65c9db61098561c48774bbe4bdb7c4a0

Airthings is the first environmental source, and it pulls CO2, radon, VOCs, particulate matter, temperature, humidity and air pressure.

The Airthings consumer API only returns the latest readings, no history, but there is a CSV upload if you have longer history you want to look at with your AI.

The point is lining it up against recovery and training. It is good for seeing how your bedroom CO2 on a given night tracked with your Nightly Recharge the next morning, or whether high temperature days match worse sleep and a tanked ANS Charge.

Airthings is in private beta for now, there is a request access button on the dashboard.

You can find it at https://freddy.coach/

It's a personal side project, and your early support is truly energizing. I know health data is sensitive and I have handled it for years with fitIQ. Data is encrypted, I do not sell it, and I am not looking to make a profit off your stats, but if you still don't trust 3rd party solutions, just don't use it :)

I use my agent to give me a morning rundown of yesterday's performance and sleep.

Just made it easier to do with a health-data MCP server, and just shipped a CLI for it. It is open source (MIT) and lets you, or a headless agent, sign in, connect wearables, and query your health data from the terminal without hand-rolling the OAuth device flow, token refresh, or any MCP plumbing.

npm i -g /cli

freddy login  # device-flow sign-in, tokens in ~/.freddy (0600)
freddy connect oura  # prints a URL to link a wearable
freddy providers  # list connectable wearables
freddy query --metrics hrv,sleep_duration --days 7
freddy metrics  # what you can query
freddy profile | freddy status
freddy sync oura | freddy disconnect oura

Every command takes --json, so an agent can shell out and parse the output directly. Access tokens last an hour and refresh automatically, so a long-running agent just keeps working.

It reads sleep, HRV, resting HR, recovery, training load, workouts, body composition and more. Connectable today for everyone: Polar, Oura, Withings, Suunto, Intervals.icu, Concept2, Wahoo, Hevy. In beta: WHOOP, Garmin, Dexcom, Strava, Ultrahuman.

The CLI is just a wrapper over freddy's MCP server, so the same data is reachable from any MCP client (Claude, ChatGPT and the rest); the CLI is for terminals and agents that prefer to shell out.

You can find it at freddy.coach, and the package is freddy-coach/cli on npm.

Hey folks, some of you know freddy already, a personal health MCP I built to solve my own problem. This community reception has been amazing, so I keep working on it. Just added Airthings environment data integration and a full CLI support with npm package.

freddy connects your health and performance data to an AI like Claude, ChatGPT, Gemini or Perplexity. You paste one personal URL into your AI client, and from then on it can read your data in any conversation. Suunto has been supported for a while, along with Oura, Polar, Withings, Intervals.icu, Concept2, Wahoo and Hevy, and more.

Airthings is the first environmental source, and it pulls CO2, radon, VOCs , particulate matter, temperature, humidity and air pressure.

The Airthings consumer API only returns the latest reading, so Freddy starts building your air-quality history from the moment you connect, and the trends get richer the longer it runs. Right now working on an import functionality, for historical data.

The point is lining it up against recovery and training. It is good for seeing how your bedroom CO2 on a given night tracked with your Body Resources the next morning, or whether high temperature days match worse sleep and slower recovery.

Airthings is in private beta for now, there is a request access button on the dashboard. It connects a bit differently to a watch login: you make a quick API client in Airthings and paste two keys, about two minutes.

https://preview.redd.it/jqchhamj2t4h1.png?width=2852&format=png&auto=webp&s=239381599d38731bf379ada7cc652d45fba6d7e1

You can find it at https://freddy.coach/

It's a personal side project, and your early support is truly energizing. I know health data is sensitive and I have handled it for years with fitIQ. Data is encrypted, I do not sell it, and I am not looking to make a profit off your stats, but if you still don't trust 3rd party solutions, just don't use it :)

Hi fellow athletes,

Quick update on Freddy, the personal MCP server I posted about a while back. You paste a URL into Claude / ChatGPT / Gemini / Perplexity and the AI can read your health and performance data in any conversation. Polar has been supported since launch, alongside Oura, Withings, Suunto, WHOOP, Garmin, etc.

Two new beta sources went live this week:

• Wahoo Cloud: ELEMNT, TICKR, KICKR, RIVAL
• Concept2 Logbook: RowErg, SkiErg, BikeErg

More sources mean richer cross-domain queries. KICKR rides and Concept2 sessions now sit next to your Polar data in the same query.

https://preview.redd.it/xxw0s6gyy31h1.png?width=533&format=png&auto=webp&s=3e07ad24e31302c803237368e45b41f7db886933

You can check it out at freddy.coach.

Currently Polar, Oura, Withings, Suunto, Intervals.icu, Hevy are available to everyone. WHOOP, Dexcom, Strava, Garmin, Wahoo, Concept2 are in beta. You can request access from the dashboard.

I know health data is sensitive and I have handled it for years with fitIQ. Data is encrypted, I do not sell it, and I am not looking to make a profit off your stats, but if you still don't trust 3rd party solutions, just don't use it :)

Hi folks,

Quick update on Freddy, the personal MCP server I posted about when Suunto went GA. You paste a URL into Claude / ChatGPT / Gemini / Perplexity and the AI can read your health and performance data in any conversation. Suunto is fully supported, alongside Polar, Oura, Withings, WHOOP, Garmin, etc.

Two new beta sources went live this week:

• Wahoo Cloud: ELEMNT, TICKR, KICKR, RIVAL
• Concept2 Logbook: RowErg, SkiErg, BikeErg

https://preview.redd.it/rvgo5o99y31h1.png?width=532&format=png&auto=webp&s=f983ecae98fbe518bcc4091481eb53d6440fd4c4

More sources mean richer cross-domain queries. Trainer rides and erg sessions now line up with your Suunto data in the same question.

You can check it out at freddy.coach.

Currently Suunto, Polar, Oura, Withings, Intervals.icu, Hevy are available to everyone. WHOOP, Dexcom, Strava, Garmin, Wahoo, Concept2 are in beta. You can request access from the dashboard.

https://preview.redd.it/tdgavzuby31h1.png?width=1120&format=png&auto=webp&s=2caaf1b35770ddcb404bc4eaac586126f393b2b6

I know health data is sensitive and I have handled it for years with fitIQ. Data is encrypted, I do not sell it, and I am not looking to make a profit off your stats, but if you still don't trust 3rd party solutions, just don't use it :)

Hi folks,

There is now a way to "ask your body anything" in your own AI (Claude, ChatGPT, Perplexity). Available data sources are broad and growing, with full Oura support, plus Hevy, Intervals.icu, Polar, Suunto, Withings, Concept2, Dexcom, Garmin, Strava, Wahoo, WHOOP. And more coming..

https://preview.redd.it/pmvtzcfjx31h1.png?width=1112&format=png&auto=webp&s=47623b24c3bb018672e2e10e3e730e2d9eaaec25

Is anyone looking for a solution like that?

Hi folks,

Quick update on Freddy, the personal MCP server I posted about when Hevy support landed. You paste a URL into Claude / ChatGPT / Gemini / Perplexity and the AI can read your health and performance data in any conversation. Hevy is fully supported, alongside Polar, Oura, Withings, Suunto, WHOOP, Garmin, etc.

Two new beta sources went live this week:

• Wahoo Cloud: KICKR, ELEMNT, TICKR
• Concept2 Logbook: RowErg, SkiErg, BikeErg

https://preview.redd.it/gf37mwxaw31h1.png?width=1119&format=png&auto=webp&s=e31a1d2cfd800dfc8d5de529a3760f5ffd5410a3

More sources mean cross-domain questions. With Hevy plus Wahoo plus Concept2 in, the AI can reason about lifts and conditioning together.

You can check it out at freddy.coach.

Currently Hevy, Polar, Oura, Withings, Suunto, Intervals.icu are available to everyone. WHOOP, Dexcom, Strava, Garmin, Wahoo, Concept2 are in beta. You can request access from the dashboard.

I know health data is sensitive and I have handled it for years with fitIQ. Data is encrypted, I do not sell it, and I am not looking to make a profit off your stats, but if you still don't trust 3rd party solutions, just don't use it :)

Hi folks,

Quick intro for newcomers and a quick update for users. Freddy is a personal MCP server I built. You paste a URL into Claude / ChatGPT / Gemini / Perplexity and the AI can read your health and performance data in any conversation. Withings has been supported since launch (Body+, Sleep Analyzer, BPM, Pulse), alongside Oura, Polar, Suunto, WHOOP, Garmin, etc.

https://preview.redd.it/gwfp4xwkv31h1.png?width=546&format=png&auto=webp&s=fde1a18a16eb219b17b73052423a4f1a6bf06673

Two new beta sources went live this week:

• Wahoo Cloud: ELEMNT, TICKR, KICKR, RIVAL
• Concept2 Logbook: RowErg, SkiErg, BikeErg

More sources mean richer cross-domain queries. Scale, sleep, and BP next to a trainer log and erg sessions in one question.

You can check it out at freddy.coach.

Currently Withings, Polar, Oura, Suunto, Intervals.icu, Hevy are available to everyone. WHOOP, Dexcom, Strava, Garmin, Wahoo, Concept2 are in beta. You can request access from the dashboard.

I know health data is sensitive and I have handled it for years with fitIQ. Data is encrypted, I do not sell it, and I am not looking to make a profit off your stats, but if you still don't trust 3rd party solutions, just don't use it :)

Hi folks,

Quick update for those of you using Freddy to look at Suunto data through Claude Desktop, ChatGPT, or Perplexity. As of this week, you can also use it from headless AI agents (OpenClaw, Claude Code, Hermes, Cursor, other CLI) on autopilot.

https://preview.redd.it/zwile3i3zg0h1.png?width=665&format=png&auto=webp&s=d3a32c942aada5e09d9b6e45d95af10719cb05cb

How it looks from a headless agent: you ask the agent to connect to Freddy, it shows you a short sign-in code and a link, you open the link once and approve, done. From that point the agent can read your Suunto data, kick off a fresh sync, or connect a new wearable, on its own schedule, on your behalf.

The reason this matters is the kind of thing it actually unlocks. A few setups I've been running:

• A morning briefing delivered to Telegram, WhatsApp, or Slack: "Nightly Recharge dropped, ANS Charge is mid, RHR is 2bpm above baseline, take it easier today."
• A daily job that pulls your Suunto data and writes it wherever you already keep notes: Notion, Obsidian, a Google Doc, your email.
• An automatic monthly report on training load, recovery, and sleep trends, summarized however you want it and sent wherever you read.
• An agent that's actually your assistant, not a chat tab. It already knows your baseline, your goals, and your context, and can act on any of them without you starting over each time.

https://preview.redd.it/lytmyyq1zg0h1.png?width=541&format=png&auto=webp&s=584a0bf46ae2bd9485973a3921e30d766bb502b6

If you're already using Freddy with Claude Desktop or ChatGPT, nothing changes for you. Your existing setup keeps working.

Site: https://freddy.coach/

I know health data is sensitive and I have handled it for years with fitIQ. Data is encrypted, I do not sell it, and I am not looking to make a profit off your stats, but if you just don't trust 3rd-party solutions, don't use it :)

Hi folks,

Quick update for those of you using Freddy to look at Polar data through Claude Desktop, ChatGPT, or Perplexity. As of this week, you can also use it from headless AI agents (OpenClaw, Claude Code, Hermes, Cursor, other CLI) on autopilot.

https://preview.redd.it/u9doyzmmi90h1.png?width=665&format=png&auto=webp&s=bd8b10e9898cd95c4e103607a35b51340497538f

How it looks from a headless agent: you ask the agent to connect to Freddy, it shows you a short sign-in code and a link, you open the link once and approve, done. From that point the agent can read your Polar data, kick off a fresh sync, or connect a new wearable, on its own schedule, on your behalf.

The reason this matters is the kind of thing it actually unlocks. A few setups I've been running:

• A morning briefing delivered to Telegram, WhatsApp, or Slack: "Nightly Recharge dropped, ANS Charge is mid, RHR is 2bpm above baseline, take it easier today."
• A daily job that pulls your Polar data and writes it wherever you already keep notes: Notion, Obsidian, a Google Doc, your email.
• An automatic monthly report on training load, recovery, and sleep trends, summarized however you want it and sent wherever you read.
• An agent that's actually your assistant, not a chat tab. It already knows your baseline, your goals, and your context, and can act on any of them without you starting over each time.

https://preview.redd.it/h94789y3j90h1.png?width=541&format=png&auto=webp&s=261cdfaf12cd9c2449f9d93e9486443df0bdd7a2

If you're already using Freddy with Claude Desktop or ChatGPT, nothing changes for you. Your existing setup keeps working.

Site: https://freddy.coach/

I know health data is sensitive and I have handled it for years with fitIQ. Data is encrypted, I do not sell it, and I am not looking to make a profit off your stats, but if you just don't trust 3rd-party solutions, don't use it :)

Hey folks,

Quite a few of you (more than I expected!) have requested Suunto access on Freddy over the last couple of weeks. As of today, the integration is officially out of beta and on production API access. So anyone can connect straight from the dashboard.

Quick context for anyone landing here cold: Freddy is a personal health-data server (an MCP server, technically) that lets your AI of choice read directly from your wearables. You paste a single URL into ChatGPT, Claude, Perplexity, Gemini, Cursor, or any other MCP-aware client, and from that point your data shows up in conversation.

https://preview.redd.it/m00ve6hxtozg1.png?width=1167&format=png&auto=webp&s=6ab0db3aacec0528e10f6ab28242eaa82072d908

For Suunto specifically, the pull covers:

• Workouts: per-session HR, pace, ascent, energy, recovery time, peak EPOC, PTE
• Routes from your activities
• Sleep: deep / light / REM, HR, HRV, SpO2
• Body Resources balance and stress state
• 24/7 heart rate, steps, energy, SpO2

So a real conversation looks like "compare my Body Resources balance for the weeks I ran 60+ km vs the easier weeks" or "I felt tired this morning, what could be the reason?" The watch knows the workout, and now the AI does too, alongside whatever else you have connected.

Polar, Oura, Withings, Hevy, and Intervals.icu are also supported. WHOOP and Dexcom CGM are still in beta (request access on the dashboard). Garmin is next on the list.

Thanks for the nudges and patience. If something breaks, ping me here or on the in-app contact form.

I log a lot of stuff on Notion. I manually log VO2max, HRV, sleep, RHR, and many more. All updated monthly copied from the app, which is roughly as fun as it sounds. I've always wanted it on autopilot, so I added that to Freddy.

Freddy is an MCP server that pipes your wearable data into any AI client that supports MCP. Notion AI is now one of them. You paste a personal URL into Notion's connector, and your Polar data become readable inside any Notion page. Build a training log that fills itself. Drop a table into your weekly review that pulls last week's sessions automatically. No exports, no .fit files, no screenshots.

https://preview.redd.it/ejumbkr7p9zg1.png?width=960&format=png&auto=webp&s=f24379b1b25d4485a0aea3489cb91305037feb77

One caveat: Notion AI only supports MCP through Custom Agents right now, so you have to wire Freddy into a custom agent rather than using the default sidebar. If there's interest, I could look into pushing the data into Notion automatically rather than going through the AI prompt. Just let me know in the comments.

Suunto, Polar, Oura, Hevy, Withings, and Intervals.icu are also supported, with WHOOP and Dexcom in beta. Garmin and Apple Health are next.

You can find it at https://freddy.coach/.

I know health data is sensitive and I have handled it for years with fitIQ. Data is encrypted, I do not sell it, and I am not looking to make a profit off your stats, but if you just don't trust 3rd-party solutions, don't use it :)