The pitch: a memory for your ChatGPT coach
GetFit MCP starts from an annoyance anyone who has used ChatGPT as a fitness coach knows. The model gives decent calorie estimates, but it forgets everything. Founder Frank Karro puts it bluntly on his own site: when he asked ChatGPT for his log history, "it invented meals I never ate." His fix is not another chatbot. It is an MCP connector that gives ChatGPT a real database to write to, so the conversation you are already having becomes the input method for a proper tracker.
You tell ChatGPT "two eggs, rye bread and a flat white, then a 40 minute run" and GetFit logs it: roughly 430 kcal and 22 g protein for breakfast, 420 kcal for the run, with a running daily total against your target. Ask "how is my week looking?" and you get adherence counts and a weight trend, computed from stored records rather than hallucinated ones.
What sits behind the chat
The part most MCP demos skip is the part Karro clearly cared about. Every log lands at app.getfitmcp.com, where you get a calorie ring, macro bars (protein against a gram target, not just calories), a weight trend chart over 30, 90 or 180 days, a steps counter and a "completeness" score for daily check-ins. Every entry keeps a full change history and is labeled by source, ChatGPT or web, so when an estimate is wrong you fix the portion size in the app instead of arguing with the model. The maintenance calorie estimate is computed from your own logs once there is enough data, which beats the generic formula number every tracker starts you with. Export is JSON or a CSV bundle, generated on demand.
The trust page is doing real work
For a one-person beta, the privacy policy is unusually specific. No third-party analytics, no tracking pixels, no ad identifiers, no model training on your data. Hosting is on Cloudflare in the EU, sign-in email goes through Resend, and ChatGPT's access runs on OAuth tokens that expire in fifteen minutes, scoped to your own account and revocable from Settings. Karro names himself and his country (Estonia) and publishes a real contact email. None of it is independently audited, but it reads like it was written by someone who expects to be held to it.
The before-and-after section is similarly concrete: Karro himself, 78 to 71.5 kg over nine months, plus two friends (98 to 75 kg in six months, 107 to 76 kg). Attributed, modest numbers, no shredded photos. I find that more convincing than the usual testimonial wall.
The gaps, and they are not small
There is no pricing. Not "pricing coming soon," just nothing: no plans, no free tier, no hint of what this costs when the beta ends. Karro mentions the custom agent he built before this cost him over 100 dollars a month, which hints at what he considers expensive but tells you nothing about what he will charge.
It is also a closed beta with an invite queue ("spots open in small batches"), so you cannot try it today. And there is no independent verification of anything. Search Reddit or Product Hunt for GetFit MCP and you will find zero threads, zero reviews, zero users talking about it. The only evidence the product works is the founder's own site. For a tool whose whole value is accumulated personal data, betting on a single-founder project with no stated business model is a real risk: if it folds, your years of logs become a JSON file and a memory.
Finally, it is ChatGPT-only. The site says nothing about Claude or any other assistant, which is odd for a product built on MCP, a protocol Anthropic created. If you do not already live in ChatGPT, this is not for you at all.
Should you join the waitlist?
If you already talk to ChatGPT about food and training and keep losing your history, this is the most sensible fix I have seen: the tracker disappears into a conversation you were having anyway, and the web app handles the parts chat is bad at. Joining the queue costs you an email address. Just treat it as an experiment, not a permanent home for your health data, until pricing exists and someone other than the founder has logged a meal with it.