Running a Cut in Claude: Setup, the Weekly Loop, and What Drifts
Published August 31, 2026
A cut is a long, boring feedback loop. Eat slightly less than you burn, wait, watch a weekly average, adjust. Almost everything that goes wrong is a measurement problem rather than a willpower problem, which is why the tracking setup matters more than the diet.
This post is the setup and the loop. It is not a case study. We have not run an eight-week cut on this stack and published the log, and where a number would normally go you will find a note saying so.
Assumed in place: the nutrition server attached to Claude Code or Cursor, a profile file with targets, and a core food list. The overview and the dietitian setup post cover both.
Set the deficit smaller than you want to
The temptation is a large deficit because it finishes sooner. The reason not to is that a large deficit makes every other part of the loop harder: hunger rises, training quality drops, adherence gets worse, and more of the lost mass is muscle.
A moderate deficit, where the commonly cited range is around 20 to 25% below maintenance and for many people that is roughly 400 to 600 kcal, is slow and boring and it works. Published guidance tends to frame the target rate as a percentage of bodyweight per week rather than an absolute, because a 60 kg person and a 110 kg person should not lose at the same rate; something in the region of 0.5 to 1% of bodyweight a week is the range usually given.
Get the maintenance estimate from the tool, then treat it as a hypothesis:
You: I am 34, male, 82 kg, 180 cm, lifting four times a week, desk job. Maintenance and a moderate cutting target please.
That is one calculate_macro_targets call. What comes back is a formula estimate scaled by an activity multiplier, not a measurement of your metabolism, and two people with identical inputs genuinely differ by several hundred calories. The formula's job is to give you a starting point to correct. The REST version of the same calculation is documented in the TDEE and macro calculator guide.
Protein is the number you do not negotiate
In a deficit, protein is doing a specific job: it is the main lever you have on keeping muscle while you lose fat, and it is the most satiating macro per calorie, which makes the diet survivable.
Published guidance for people lifting in a deficit clusters somewhere around 1.6 to 2.2 g per kg of bodyweight, and the upper end of that range is where recommendations tend to sit when the deficit is aggressive or the person is already lean. Pick a point in the range, write it in the profile file, and hold it constant while you adjust calories. Changing two variables at once tells you nothing about either.
Practically, hitting a high protein target at a reduced calorie intake is a constraint-satisfaction problem, which is exactly what a catalog-backed planner is good at. Order the plan protein-first, as in meal planning with Claude. Calories are easy to add at the end, and wedging 175 g of protein into a budget you already spent on carbohydrate is not.
The weekly loop
Daily logging, weekly decisions. The distinction matters, because daily weight is mostly water and glycogen and reacting to it will make you miserable and worse off.
Every day: log food as you eat it, and weigh yourself in the same conditions: waking, after the toilet, before food. Two lines in the log file.
Every week, one review:
You: Read the last 14 days of the log. Weekly average weight for each week, average calories, average protein, days logged out of 7. Then tell me whether to hold or adjust.
Four numbers come out of that and each one is a different decision.
Days logged. Look at this first. If it says four out of seven, nothing else in the review means anything, because the average is over the days you chose to record and those are systematically the good days. Fix adherence before you touch calories.
Average calories versus target. If actual is 200 above target, the deficit is not the size you think it is. That is not a metabolism finding, it is an arithmetic one.
Average protein. If it is short, look at which days. Training days are the usual culprit.
Weekly average weight versus last week. Only now does this number mean anything. Compare week averages, never days. And expect noise: a single flat week in the middle of a well-executed cut is common and is not evidence of anything.
Telling portion creep from adaptation
Four weeks in, the scale stops moving. There are two candidate explanations and they call for opposite responses, so distinguishing them is the highest-value diagnostic in the whole process.
Portion creep. Your logged intake is accurate and your actual intake has drifted upward. Oil pours get bigger, "a handful" grows, the thing eaten standing at the counter does not get logged, the weekend is estimated rather than weighed. This is overwhelmingly the more common explanation, and the research on self-reported intake is fairly unanimous that people under-report, often substantially, without intending to.
Adaptation. Your expenditure has fallen, some from carrying less mass and some from moving less spontaneously as energy drops. Real, and smaller than the internet believes.
Here is how to tell. Weigh everything for four days. Everything, including oil, including the bite off someone else's plate, including the milk in coffee:
You: Strict mode for four days. I will give you weights for everything. Flag anything I log without a weight.
If the logged total comes out higher than your usual logged total, it was creep, and the fix is accuracy rather than a lower target. If four strictly weighed days land on the number you have been logging all along, it is adaptation, and you drop the target by 100 to 150 kcal or add activity.
Guessing between the two and cutting calories anyway is how people end up eating very little and still not losing.
Four places a cut quietly goes wrong
The weekend. Two estimated days out of seven is 28% of the week, and estimated days run low. If Saturday is the one day you do not weigh, the cut is happening five days a week. Log the weekend to the same standard, or accept that your weekly average is a fiction.
Liquid calories. Milk-based coffees, juice, alcohol at roughly 7 kcal per gram, the oil in a salad dressing. None of them feels like eating. All of them count.
Foods with a health halo. Nuts, olive oil, nut butter, avocado, granola, dark chocolate. Every one is calorie-dense and every one is the sort of thing people log by eye rather than by weight. Thirty grams of nuts and what you actually poured are usually different amounts.
Cooking fat. Sixty grams of olive oil is over 500 kcal. It goes in the pan, not the bowl, so it does not feel like food. This is the single most commonly omitted item in any food log, and recipe macros without a spreadsheet covers why it distorts a whole recipe.
Diet breaks, and stopping
A long cut does not have to be continuous. Planned periods back at maintenance, a week or two at a time, make a long phase more survivable and are a normal part of how coaches structure things. Put them in the plan in advance rather than taking them by accident.
Have an exit condition written down before you start. A goal weight, a body-fat estimate you are comfortable with, or a duration. Cuts without an end condition turn into a permanent state, and permanent restriction is not a good place to live.
One thing worth writing into the plan alongside the exit condition: what happens afterwards. The end of a cut is where most of the regained weight comes from, because there is no plan for the week after. Deciding in advance that you return to maintenance for a month before doing anything else removes that decision from a moment when you are hungry and pleased with yourself.
Stop early, without negotiation, if training performance collapses, sleep goes, you are thinking about food constantly, or eating starts to feel fraught. Those are not signs to push harder. If there is any history of disordered eating, do not run a self-directed cut at all. That needs a clinician involved, and a calorie total does not capture what is being managed.
What the tracking setup does not do
It does not supply adherence. Nothing here makes you eat the food or log the day. The published work on food journaling is consistent that consistency beats precision, which means a rough log you keep for four months beats an immaculate log you keep for nine days.
It does not know your context. No labs, no medication, no sleep data, no menstrual cycle. That last one materially affects water weight and therefore the weekly average, which is a good argument for comparing like phases rather than adjacent weeks.
It does not keep your history. The server has no logging tool. Your log is a file you own, which is a limitation and also the reason you can read it in five years.
Connect it
Claude Code:
claude mcp add --transport http calorie-api \
https://calorieapiadmin.com/mcp \
--header "X-API-Key: YOUR_KEY"
Cursor:
{
"mcpServers": {
"calorie-api": {
"url": "https://calorieapiadmin.com/mcp",
"headers": { "X-API-Key": "YOUR_KEY" }
}
}
}
Browser sign-in for Claude apps is off on this server until OAuth is enabled, so those are the two clients. The MCP page, the docs and MCP pricing have the overview, reference and plan.
A logged day is 12 to 16 tool calls, and the weekly review is zero because it reads the file. Against 20 a minute and 10,000 successful calls a month, a full cut fits comfortably. On the 7-day trial, which is 5 a minute and 200 calls total that never reset, you get roughly a fortnight of careful logging, which is long enough to decide whether the workflow suits you and not long enough to run a cut.
What we have not measured
We have not run an eight-week cut on this setup and published the log. No adherence rate, no weekly drift figures, no comparison between logged and weighed intake, no measured outcome of any kind.
Everything above is either mechanism (what the tools do, what the limits are, how the loop works) or published general guidance about deficits and protein, which is other people's work and is presented as ranges rather than as prescriptions. When we run a cut on this stack, the log and the method will be published together, including the weeks it went wrong.
Related
Frequently Asked Questions
How big should the deficit be?
Smaller than you want. Commonly cited guidance is around 20 to 25% below maintenance, with a target rate framed as roughly 0.5 to 1% of bodyweight a week rather than a fixed number of kilos. A large deficit makes hunger, training quality and adherence all worse and costs more muscle.
How do I know whether a stall is portion creep or adaptation?
Weigh everything strictly for four days. If the logged total comes out higher than your usual logged total, it was creep and the fix is accuracy. If four strict days land on the number you had been logging, it is adaptation and you lower the target by 100 to 150 kcal or add activity.
Which number should I look at first in a weekly review?
Days logged out of seven. If it is four, nothing else in the review means anything, because the average covers the days you chose to record and those are the good days. Fix adherence before touching calories.
Why weekly averages instead of daily weight?
Daily weight is mostly water and glycogen. Reacting to it produces bad decisions and a miserable time. Compare one week's average to the previous week's average, and expect a flat week in the middle of a well-run cut.
Is this an eight-week case study?
No. We have not run an eight-week cut on this setup and published the log, so there are no adherence rates, drift figures or outcomes here. The post is mechanism plus published general guidance presented as ranges.