Nutrition MCP

Three steps. No app to learn.

1 / 3

Connect once

Works with any AI client that supports remote MCP servers — Claude, ChatGPT, and more. No install, no API keys.

2 / 3

Just say what you ate

Describe it in plain language — or send a photo of your meal, a screenshot from a delivery app, or a barcode (it looks the product up in Open Food Facts). Generic foods use USDA FoodData Central values when available. Macros logged automatically.

3 / 3

Track & review

Ask for daily summaries, weekly trends, goal progress, or export everything you've logged as CSV files — completely free.

Connect in under a minute

Works with any MCP client that supports OAuth 2.0 with PKCE. On first connect you create an account with Google or an email and password; sign in the same way to keep your data.

https://nutrition-mcp.com/mcp

Choose your AI client

Claude ChatGPT Other agents

Add to Claude

On the directory page, click Connect, then continue with Google or sign in with an email and password.
Done. It works right away and shows up in your iOS and Android apps automatically.

Works on every Claude plan, the free one included. To add it by hand instead, use Customize → Connectors → Add custom connector with https://nutrition-mcp.com/mcp.

Open ChatGPT on the web → Settings → Apps.
Click Create app at the bottom of the popup. If you don't see it, turn on Developer mode in Advanced settings.
Give it a name, for example Nutrition.
For Connection, paste https://nutrition-mcp.com/mcp.
For Authentication, choose OAuth — leave everything else as it is.
Check "I understand and want to continue".
Click Create.
Click Sign in with Nutrition — the login page opens; continue with Google or sign in with an email and password.
Done. It works right away and shows up in your iOS and Android apps automatically.

{
  "mcpServers": {
    "nutrition": {
      "url": "https://nutrition-mcp.com/mcp"
    }
  }
}

Add the config above to your client (Cursor, VS Code, Claude Code, and more). Windsurf uses serverUrl instead of url. In Claude Code, run claude mcp add --transport http nutrition https://nutrition-mcp.com/mcp. Your client handles the OAuth login automatically.

Set up once — or just start talking

This is completely optional — Nutrition MCP works the moment you connect. If you want, these three quick steps make it more accurate, but you can skip straight to logging.

01
Set your timezone — so days roll over at your local midnight and today's totals stay accurate wherever you are.

Just say Set my timezone to New York
02
Set your goals — daily calorie, macro, and water targets, plus an optional target weight and your preferred weight unit (kg or lb), to track your progress against.

Just say Set my daily goal to 2,000 calories and 150g of protein
03
Set your language — the language in-chat widgets (dashboards, charts) are shown in, not what the AI writes back to you.

Just say Show my widgets in German
04
Start logging — just say what you ate, send a photo or scan a barcode. That's it.

Just say I had oatmeal with berries for breakfast

Everything here is optional. You can do it now, later, or never — just start logging and set these whenever you like.

Curious what it can actually do?Browse all 48 tools — logging, barcodes, water, weight and body measurements, goals, and trends — with a description and an example prompt for each. Explore the tools

Just talk to it.

A few of the things you can do — just by talking.

Log in plain words

Say it the way you'd tell a friend. If the portion or which meal it was is missing, it asks before estimating. Fiber, sugar and added sugar go on every entry, caffeine on anything that has some, and “around 7” lands at 07:00 in your timezone.

log_meal

Saves breakfast with macros, fiber, sugar, added sugar and the coffee's caffeine

Also uses

log_water

Records the 500 ml at 07:00 toward today's water total

get_current_time

Works out today's date so “around 7” lands at 07:00

oatmeal with blueberries and a coffee for breakfast

How big was the bowl, and was the oatmeal made with milk or water? Anything in the coffee?

normal bowl, made with milk. coffee was black

Logged breakfast: about 320 kcal, 11 g protein and 6 g fiber. The coffee puts you at 95 mg of your 400 mg caffeine limit.

oh and i drank like half a liter of water around 7, before my run

Added 500 ml of water at 07:00 this morning.

Snap your plate

A photo starts a conversation, not a guess. It asks whether the meal was homemade or from a restaurant, checks your past logs for what the photo can't show, asks how much you ate in plain measures, and logs only once you approve the summary.

search_meals

Finds your earlier borscht at Tsybulya and what was in it

Also uses

log_meal

Logs it with the restaurant's name, once you say yes

lunch

Looks like borscht with sour cream and a slice of rye bread. Homemade, or from a restaurant? If it was a restaurant, which one?

Tsybulya, on Podil in Kyiv

You've logged borscht at Tsybulya twice before, both times beef with sour cream and a slice of rye, about 520 kcal. Same again? And did you finish the bowl and the bread?

same, all of it, bread too

So: beef borscht with sour cream (1 bowl, finished) and rye bread (1 slice) at Tsybulya, Podil, Kyiv. Same as last time: about 520 kcal and 24 g protein, estimates rather than published figures. Log it?

yes

Logged. That's 520 of your 2,000 kcal for today, 1,480 to go.

Scan a barcode

Photograph the package: the digits under the barcode are looked up in Open Food Facts, and it asks how much you had. Added sugar is read where Open Food Facts has it. Open Food Facts carries no caffeine, so that comes from the label or typical amounts; a missing fiber, sugar or added-sugar figure is estimated rather than left blank.

lookup_barcode

Pulls the pot's label figures and Nutri-Score from Open Food Facts

Also uses

log_meal

Logs the whole pot as breakfast, scaled from the per-100 g figures

had this for breakfast. is it any good?

That's a 150 g pot of plain Greek yogurt (0% fat): 87 kcal, 15 g protein and 5 g sugar, none of it added, and Open Food Facts rates it Nutri-Score A. Did you have the whole pot?

whole pot

Logged as breakfast: 87 kcal, 15 g protein and 5 g sugar. 145 g of protein to go today.

Generic foods from USDA

Name a plain food and your AI looks it up in USDA FoodData Central, then scales the per-100 g values to the grams you ate. Anything that record lacks is estimated rather than shown as zero, and every value is tagged with its source.

search_foods

Finds the USDA entries for “grilled chicken breast”

Also uses

get_food_macros

Scales the USDA values to 150 g

log_meal

Logs the lunch, each value tagged USDA or estimated

150 g grilled chicken breast for lunch

Logged lunch: 248 kcal, 46.5 g protein and 5.4 g fat. The protein and fat come from the USDA FoodData Central record for cooked chicken breast, scaled to 150 g. Added sugar isn’t in that record, so it is estimated.

Save your usuals

Save a meal you eat often, with its ingredients, and log it again by its name. Leave an ingredient out or change its amount just for that time, or log more than one serving; the saved meal stays as it was.

log_saved_meal

Logs the bowl without the sour cream and with the chicken doubled to 240 g

Also uses

save_meal

Keeps the bowl's six ingredients and values for one serving, without adding it to today

save my burrito bowl so i can log it fast: 150 g rice, 120 g chicken, 100 g black beans, 60 g salsa, 30 g cheese, 30 g sour cream

Saved “Burrito bowl” with its 6 ingredients, one serving: 725 kcal, 59 g protein, 22 g fat. Saving doesn't add it to today — just name it when you've had it.

had the burrito bowl for lunch. no sour cream today and double chicken

Logged lunch from your Burrito bowl, without the sour cream and with 240 g of chicken: 865 kcal and 95 g protein. The saved bowl stays as it was for next time.

where's most of that protein from?

The chicken: 74 of the 95 g. Black beans add 9 g, cheese 7 g, rice 4 g and salsa 1 g.

Set goals, check in

Set targets to reach, like calories and protein, and limits to stay under for added sugar, saturated fat, total sugar, alcohol and caffeine; only what you mention changes. Then ask about any day to see what's left to reach each target and where you stand against each limit.

set_nutrition_goals

Saves calorie and protein targets and a 400 mg caffeine limit

Also uses

get_goal_progress

Shows today against each target and limit in a card in the chat

trying to lean out a bit. set me at 2000 cals and 160g protein. also what's a sane caffeine cap? i basically live on coffee

For healthy adults, EFSA and the FDA put it at up to 400 mg a day, roughly four cups of brewed coffee, and EFSA advises 200 mg in pregnancy. The limit is your call; want 400?

yeah 400

Done: 2,000 kcal and 160 g protein to reach, caffeine under 400 mg. Everything else stays as it was: 220 g carbs, 70 g fat, 20 g saturated fat, 30 g fiber, 2.5 L water, your 25 g added-sugar limit and your 75 kg target weight.

can i still squeeze in dessert tonight?

You're at 1,540 of 2,000 kcal and 104 of 160 g protein, so 460 kcal and 56 g protein to go. Added sugar is at 16 g of your 25 g limit, and caffeine at 130 of 400 mg, both under.

Review the week

Ask how the week went and the trends card opens in the chat with your average calories and macros against your goals and how many days you logged. Switch between 7, 14 and 30 days without a new request. It can also tell you your streaks and days on target.

get_trends

Averages your last 7, 14 and 30 days and opens the trends card

how did the last two weeks go? feels like i've been all over the place

You logged 13 of the 14 days, the last 10 in a row, and 9 of them landed within 10% of your 2,000 kcal target. Calories averaged 1,830 a day, but that counts the day you skipped as zero.

what about sugar?

Added sugar averaged 22 g a day, under your 25 g limit, but went over it on 5 of the 13 days. Total sugar averaged 52 g, mostly from fruit and milk.

Track weight and measurements

Log a weigh-in in kg or lb and see the trend under the day-to-day noise: a smoothed average, the change over the month and the gap to your target. Tape measurements of nine body sites, waist to calf, sit alongside it in cm or in. An impossible reading, like a missed decimal point, is caught before it's saved.

log_weight

Saves this morning's 78.4 kg weigh-in

Also uses

get_weight_trends

Smooths day-to-day swings into a trend weight and weekly rate, with the gap to target

log_body_measurement

Saves the 84 cm waist measurement

get_body_measurements

Finds the last waist entry to compare against

78.4 kg this morning. how's my trend looking?

Saved 78.4 kg. You're down 1.8 kg since 11 Feb (from 80.2 kg), and your 7-day average is 78.7 kg. 3.4 kg to go to your 75 kg target.

waist was 84 cm too. how does that compare?

Saved your waist at 84 cm. Your last one was 86.5 cm on 14 Feb, so that's 2.5 cm less.

Spot your patterns

Ask when you eat more and it compares your own logs: weekdays with weekends, days with and without breakfast, big lunches, early dinners with late ones. It flags unusual days too, and reports patterns, not causes.

get_meal_patterns

Compares breakfast, dinner-time and weekend days over 30 days

do i eat more on days i skip breakfast? feels like it

A little, yes. You logged 28 of the last 30 days and skipped breakfast on 6. Those days averaged 2,215 kcal and 88 g protein, against 1,940 kcal and 123 g on the 22 days with breakfast.

what about late dinners? on gym days i don't eat till 9

Those stand out too: the 9 days with dinner at 8 pm or later averaged 2,165 kcal, against 1,920 on the other 19. Both are patterns in your log, not proof either one causes it.

and weekends?

Barely different: weekends average 2,010 kcal and weekdays 1,993.

Track drinks too

Alcohol stays hidden until you switch tracking on. Pick US standard drinks or UK units, and each drink is worked out as grams of alcohol from its size and strength. Switching it off later hides alcohol again; it doesn't delete it.

set_alcohol_tracking

Turns alcohol on, shown in grams alongside UK units

Also uses

log_meal

Logs the pint with its alcohol as grams of ethanol (568 ml × 4% × 0.789)

can you count my drinks too? trying to cut back a bit

Sure. Should I show alcohol as US standard drinks or UK units?

uk units, i'm in manchester

Alcohol tracking is on, in grams alongside UK units.

had a pint of lager with dinner

What strength was it? Lagers usually run 4 to 5%, and the tap or can will say.

4% i think

Logged as a dinner entry: a 568 ml pint at 4% is 17.9 g of alcohol, 2.3 UK units, about 180 kcal.

Bring your history

Moving from MyFitnessPal, Cronometer, Lose It! or MacroFactor? The importer opens in the chat, reads and maps your export in the browser, saves nothing until you confirm a preview, and doesn't double a meal if you import the same file again with the same timezone. If your timezone isn't set, that's sorted out first, so every meal lands on the right day.

get_profile

Shows the account has no timezone set yet

Also uses

set_timezone

Sets America/Chicago so meals land in local time

start_meal_import

Opens the in-chat importer for your export file

bulk_import_meals

Saves the confirmed rows in batches of up to 50, skipping any already imported

switching over from myfitnesspal. can you bring my food log across? i've got the export

Before I open the importer: your account has no timezone set, so the dates in your export would be read as UTC and meals could land on the wrong day. Want to set it first? Which timezone are you in?

chicago

Timezone set to America/Chicago, and the importer is open above: pick your MyFitnessPal export file there. It's read right in your browser, and nothing is saved until you confirm the preview.

it matched the columns but food name says not in this file?

That's normal for MyFitnessPal: its export has one row per meal per day, with totals but no food names. Rows without a food name are labelled with the app they came from, and their calories and macros come through as they are.

603 meals. it says they'll be logged at midday?

Right, the export has dates but no meal times, so each one is placed at local noon. That keeps every meal on the right day in Chicago; you just won't see when you actually ate.

done! did everything make it?

All of it: 603 meals imported and none failed.

Take your data

One request packs everything stored about you — meals, saved meals and their ingredients, water, weigh-ins, body measurements, goals, profile settings, your account record, usage telemetry and app connections — into a ZIP of CSV files with a README. Timestamps use your timezone, and meals.csv imports straight back in.

export_all_data

Zips every CSV file and a README behind a 60-minute link

my dietitian wants to see my food log. can you get it out as a spreadsheet?

Done. The ZIP has 1,204 meals, 318 water entries and 64 weight entries, plus your nutrition goals, profile settings, account details, tool-usage telemetry and app connections. Your food log is meals.csv inside.Download your export (link valid for 60 minutes)

Breakfast somewhere, dinner somewhere else.

Live nutrition stats across every Nutrition MCP account — calories, food logs, macros and weight lost — refreshed every five seconds.

Refreshes every 5 s · next in 5sLive since you opened this page · 0:00

—Calories logged

—Food logs

—Protein logged

—Carbs logged

—Fat logged

—Weight lost since July 2, 2026

—Water logged

— timezones · days roll over at each person's own midnight dot size = share of profiles

Totals across every account, updated as meals are logged. Individual data is never shown.

What you can track

Meals in plain language

Describe what you ate — generic foods get USDA FoodData Central values when available, and your AI estimates the rest: calories, protein, carbs, fat, saturated and trans fat, fiber, total sugar, added sugar, and caffeine in milligrams. It logs the meal and tags each value with its source.

Scan a barcode

Snap or type a product barcode and pull macros, fiber, and sugar from Open Food Facts — added sugar too, where it's listed — scaled to how much you ate.

Goals & progress

Set daily calorie, macro, fiber, and water targets — plus added sugar, saturated fat, total sugar, caffeine, and alcohol limits to stay under — and check live progress toward them.

Summaries & trends

Daily and weekly breakdowns, 7/14/30-day trends, streaks, and recurring meal patterns.

Water logging

Track hydration in milliliters alongside your meals and review it by day.

Weight tracking

Log your body weight in kg or lb, see a smoothed trend weight and weekly rate across your whole history, and track progress toward a target weight. Tape measurements of nine body sites — waist to calf — in cm or in sit alongside it.

Timezone-aware

Days roll over in your local time, wherever you are in the world.

Import from another app

Bring your meal history over from MyFitnessPal, Cronometer, Lose It!, or MacroFactor — or any other CSV, by mapping its columns yourself. You confirm what gets added before anything is saved.

Export & own your data

Take everything we store about you — meals, saved meals and their ingredients, water, weight, body measurements, goals, and profile, plus your account record, usage telemetry, and connected apps — as one ZIP of CSV files. Meals are the only part that can be imported back in for now. Delete your account and data whenever you want.

Talking beats tapping.

Snap a barcode or just say what you ate — no need to look up each food yourself, no separate app to open.

Traditional apps

Search a database for every item
Fix wrong database entries by hand
Yet another app to open, often behind a paywall
Tedious manual logging

Describe meals in plain language
USDA values for generic foods, estimates otherwise
Works inside Claude or ChatGPT, free
Ask for trends, summaries, and goals

Switching from a specific app? See how Nutrition MCP compares to MyFitnessPal, Cronometer, and other trackers.

Private by defaultNever sold, shared or used for ads.

Open sourceAudit or self-host it.

Export anytimeEverything we store, as CSV in one ZIP.

Delete instantlyRemove your account & data.

Help keep it running.

Nutrition MCP is free and ad-free. Patreon covers the server and database bills.

Free member

$0

Follow along — get news and updates about the server, new tools, and what's coming next.

Follow on Patreon

Paid member

Pay what you want

If Nutrition MCP is useful to you, you can help with its hosting and database costs. Everyone gets the same features, supporters included, and it stays free for all.

Become a supporter Star on GitHub

Start tracking in under a minute.

Free and open source — it works with the AI you already use.

Quick install Star on GitHub

Questions or feedback?

Found a bug, want a feature, or just have a question? Email me directly — I read every message.

anton@nutrition-mcp.com Send an email

Frequently asked questions

What is Nutrition MCP?

Nutrition MCP is a free, open-source Model Context Protocol (MCP) server that turns Claude, ChatGPT or another MCP client into a calorie and macro tracker. Instead of looking up each item yourself, you tell your AI what you ate and it logs the calories, macros, fiber, sugar, added sugar, saturated and trans fat and caffeine to your own food diary.

What is the Model Context Protocol (MCP)?

The Model Context Protocol is an open standard that lets AI assistants like Claude and ChatGPT connect to external tools and data sources. An MCP server provides specific capabilities — here, nutrition tracking — that the AI can use during a conversation. Think of it as a plugin system for AI assistants.

How do I track calories with Claude or ChatGPT?

Connect Nutrition MCP once — in Claude from the connectors directory, in ChatGPT as a custom app with the server URL — and sign in. Then tell your AI what you ate in your own words, show it a photo of the meal, or give it a product barcode. For a generic food, the values come from USDA FoodData Central when it has them and are estimated otherwise; a barcode is looked up in Open Food Facts. Nutrition MCP saves the entry to your food diary. Ask for today's totals, weekly trends or progress toward your goals at any time.

Does it work with ChatGPT?

Yes. In ChatGPT on the web, open Settings → Apps, create a custom app with the server URL using OAuth, and sign in. Creating a custom app needs ChatGPT's Developer mode, which OpenAI offers on some ChatGPT plans.

Which other clients are supported?

Any MCP client that supports OAuth 2.0 with PKCE — including Claude.ai, the Claude desktop and mobile apps, Claude Code, Cursor, Windsurf, and VS Code.

Can I self-host it?

Yes. Nutrition MCP is open source (MIT). You can run your own instance with your own Supabase project — the GitHub repository includes a full self-hosting guide and a Dockerfile.

Is Nutrition MCP free?

Yes, it is completely free — no paid tier, no ads, no hidden costs. You need an AI app that supports MCP connectors, such as Claude or ChatGPT, and a free Nutrition MCP account, which you create the first time you connect. Voluntary donations on Patreon help cover server costs and unlock nothing.

What can I track?

Calories, protein, carbohydrates, fat, saturated and trans fat, fiber, total sugar, added sugar, and water for every entry — described in plain language, looked up as a generic food in USDA FoodData Central, or pulled from a product barcode via Open Food Facts. A meal can be logged ingredient by ingredient, each with its own amount and nutrients, and a meal you eat often can be kept as a saved meal and logged again by its name. Caffeine is tracked too, in milligrams, the unit every label uses, and it adds no calories. Alcohol can be tracked as well, in grams of pure ethanol; it is shown once you switch alcohol tracking on. You can also log your body weight in kg or lb and track trends toward a target weight. Body measurements (waist, hips, neck, chest, shoulders, upper arm, forearm, thigh and calf) can be logged in cm or inches too. View daily summaries, query meals by date range, update or delete past entries, set goals, and monitor trends over time.

Can I save meals I eat often, or my own recipes?

Yes. Ask to save a meal under a name — just its totals, or a recipe with its ingredients — and its values for one serving are kept. Next time, name it and it is logged in one step: scaled by servings, with a single ingredient set to the amount you actually had or left out. The logged entry is a copy, so editing or deleting the saved meal later never changes meals already logged from it. Saved meals are part of your data export.

How accurate are the calorie counts?

Generic foods use USDA FoodData Central values when available, and estimates otherwise. A barcode's figures come from the product's record in Open Food Facts, scaled to how much you had. Each value is tagged with its source (USDA, Open Food Facts, you or an estimate), and a USDA or Open Food Facts tag appears only when the figure matches that record for the amount you logged. Estimates and records can be wrong, so check anything that matters — you can correct or delete any entry just by asking. Nutrition MCP is a logging tool, not medical or dietary advice: talk to a doctor or dietitian before making decisions about your health, especially if you are pregnant, have a medical condition or a history of disordered eating.

Does it track alcohol?

Yes, as an opt-in: alcohol tracking is off by default, and alcohol stays hidden from your meals, goals, and summaries until you switch it on. Then drinks are shown in grams of pure ethanol and as US standard drinks or UK units, whichever you prefer. Nothing infers alcohol for you: it is only recorded from a drink you log or an alcohol column in a file you import, and a drink you log is saved even while tracking is off. Switching it back off hides alcohol again and stops the importer reading alcohol columns — it is not a delete switch, and your export always includes what you logged. To remove an alcohol figure, delete the meal it belongs to.

Can I import my history from MyFitnessPal or another app?

Yes. Ask to import your history and an importer opens in the chat: you pick the CSV your old app exported, check how its columns map, and see what will be added before confirming. Exports from MyFitnessPal, Cronometer, Lose It!, and MacroFactor are recognised automatically, and any other CSV works by mapping the columns yourself. Your browser reads the file, so the AI never retypes your rows. In clients without in-chat panels you can paste your export instead — and importing the same file again does not create duplicates, as long as your timezone hasn't changed in between.

Is my data private?

Your logs are stored in the EU and linked to your own account, which you reach through the AI apps you connect. Nutrition MCP never sells your data, never shares it with third parties, apart from the food names it sends to USDA FoodData Central or Open Food Facts to look a food up, and never uses it for advertising; the home page shows only anonymous site-wide totals. Whatever your AI reads through the tools is sent to that AI's provider under your own agreement with them. You can export everything we store about you, or delete your account and all its data, at any time — the privacy policy has the details.

Track your nutrition by talking to your AI.

Three steps. No app to learn.

Connect once

Just say what you ate

Track & review

Connect in under a minute

Set up once — or just start talking

Just talk to it.

Log in plain words

Snap your plate

Scan a barcode

Generic foods from USDA

Save your usuals

Set goals, check in

Review the week

Track weight and measurements

Spot your patterns

Track drinks too

Bring your history

Choose your export

Take your data

Breakfast somewhere, dinner somewhere else.

What you can track

Meals in plain language

Scan a barcode

Goals & progress

Summaries & trends

Water logging

Weight tracking

Timezone-aware

Import from another app

Export & own your data

Talking beats tapping.

Traditional apps

Nutrition MCP

Help keep it running.

Free member

Pay what you want

Latest from PatreonFree

Start tracking in under a minute.

Questions or feedback?

Frequently asked questions