MyFitnessPal MCP Server

A Model Context Protocol (MCP) server that enables AI assistants like Claude to interact with your MyFitnessPal data, including food diary, exercises, body measurements, nutrition goals, and water intake.

Features

How diary writes work

MyFitnessPal has no public API. Reads here are scraped from the website via python-myfitnesspal, and diary writes go through MFP's internal v2 JSON API — the same one their web client uses, authenticated with your existing session token.

Custom-food writes (mfp_create_custom_food, mfp_list_own_foods, mfp_delete_custom_food) use a different endpoint family: MFP's v2 API exposes no custom-food create, so these call the same cookie-authenticated web endpoints their website uses (/api/auth/csrf, /api/services/users/foods/mine, /api/services/foods). No browser needs to be running — the stored session cookies are sufficient.

That interface is undocumented and was determined by observing the web client. It works today, but MyFitnessPal can change it without notice. They have already done so once: this server originally posted to /food/diary/{user}/add, which now returns 404, leaving food logging broken. If logging starts failing, that is the most likely cause.

How fasting works (write-only)

Fasting tools (mfp_log_fast, mfp_update_fast, mfp_delete_fast) POST / PATCH / DELETE against /v2/diary/fasting_entry on the same v2 JSON API. Payloads match the shape captured from the iOS app:

{"items": [{
  "type": "fasting_entry",
  "id": "UPPERCASE-UUID",
  "fast_started": "2026-08-06T13:00:00Z",
  "fast_ended":   "2026-08-07T05:00:00Z"
}]}

Ids are client-generated UUIDv4s in uppercase (the iOS convention). The MCP auto-generates one when you omit id from mfp_log_fast; save the returned id if you plan to update or delete the entry.

There is no read endpoint. MFP exposes no GET for fasting entries — GET /v2/diary/fasting_entry returns 405 Method Not Allowed. The mobile app populates its Fasting History screen via a delta-sync channel (mobile-sync-api.myfitnesspal.com/v2.1/sync) that requires a pre-issued sync token, is scoped to the mobile OAuth client, and rejects the web-session bearer this server authenticates with. You'll continue to read fasting history in the MFP app itself until Under Armour publishes something.

The write tools are still useful for automating log-entry (e.g. inferring fasts from your Garmin sleep window + first-meal timestamp) or correcting entries programmatically.

Prerequisites

• Python 3.10–3.12 (check with python3 --version)

Not 3.13+: lxml, pulled in by myfitnesspal, has no wheels for it and fails to build against the 3.14 C API. On macOS, brew install [email protected].

• pip 21.3+ (for pyproject.toml support; upgrade with pip install --upgrade pip)
• MyFitnessPal account
• One of the following for authentication:
• Recommended (macOS): any Chromium-based browser (Arc, Chrome, Edge,

Brave, Vivaldi, Opera, ...) with an active MyFitnessPal login session — the MCP auto-discovers the session on next call

• Firefox with an active MyFitnessPal login session (via the

browser_cookie3 fallback)

• Legacy: your MFP username/email and password (see caveats below —

MFP's NextAuth backend rejects the form-POST flow, so credential auth only works while cached cookies remain valid)

Authentication Options

This MCP supports multiple authentication methods:

Note: MyFitnessPal migrated their authentication to NextAuth, so the legacy form-POST authenticate_with_credentials path almost always fails for fresh logins. The Chromium auto-discovery path is the reliable way to get a session on macOS — just log in via any modern browser and the MCP picks it up automatically on the next call.

Installation

Option 1: Install from Source (Recommended)

# Clone the repository
git clone https://github.com/YOUR_USERNAME/myfitnesspal-mcp-python.git
cd myfitnesspal-mcp-python

# Create virtual environment (use python3.10+ on macOS/Linux)
python3 -m venv venv
# On macOS, you may need to specify version: python3.12 -m venv venv

# Activate virtual environment
source venv/bin/activate  # macOS/Linux
# On Windows: .\venv\Scripts\activate

# Upgrade pip (required for pyproject.toml support)
pip install --upgrade pip

# Install the package in editable mode
pip install -e .

Option 2: Install with pip (when published)

pip install mfp-mcp

Note: Option 2 requires the package to be published to PyPI. For now, use Option 1.

Verify Installation

After installation, verify the server can start:

# With venv activated
python -m mfp_mcp.server

You should see the server waiting for input (it communicates via stdio). Press Ctrl+C to stop.

To test authentication (optional):

MFP_USERNAME="your_email" MFP_PASSWORD="your_password" python -c "
from mfp_mcp.server import get_mfp_client
client = get_mfp_client()
print('Authentication successful!')
"

Configuration for Claude Desktop

Step 1: Locate Your Config File

Step 2: Add the MCP Server Configuration

If the file doesn't exist, create it. Add or merge the following configuration:

#### Option A: With Encrypted Credentials (Enhanced Security)

Encrypt your credentials before storing them in the config file. See Encrypted Credentials for setup instructions.

⚠️ Security note: Encryption only provides meaningful protection if MFP_SECRET_KEY is stored separately from the config file (e.g., set in your shell profile or OS keychain). Storing the key alongside the encrypted values in the same config file means anyone who obtains the config can still decrypt your credentials.

macOS Example (with key set separately in your shell environment):

{
  "mcpServers": {
    "myfitnesspal": {
      "command": "/Users/yourname/myfitnesspal-mcp-python/venv/bin/python",
      "args": ["-m", "mfp_mcp.server"],
      "env": {
        "MFP_USERNAME": "gAAAAAB...<encrypted_email>",
        "MFP_PASSWORD": "gAAAAAB...<encrypted_password>"
      }
    }
  }
}

#### Option B: With Plain Credentials (No Browser Required)

macOS Example:

{
  "mcpServers": {
    "myfitnesspal": {
      "command": "/Users/yourname/myfitnesspal-mcp-python/venv/bin/python",
      "args": ["-m", "mfp_mcp.server"],
      "env": {
        "MFP_USERNAME": "[email protected]",
        "MFP_PASSWORD": "your_password"
      }
    }
  }
}

Windows Example:

{
  "mcpServers": {
    "myfitnesspal": {
      "command": "C:\\Users\\YourName\\myfitnesspal-mcp-python\\venv\\Scripts\\python.exe",
      "args": ["-m", "mfp_mcp.server"],
      "env": {
        "MFP_USERNAME": "[email protected]",
        "MFP_PASSWORD": "your_password"
      }
    }
  }
}

#### Option C: Without Credentials (Browser Cookie Fallback)

macOS Example:

{
  "mcpServers": {
    "myfitnesspal": {
      "command": "/Users/yourname/myfitnesspal-mcp-python/venv/bin/python",
      "args": ["-m", "mfp_mcp.server"]
    }
  }
}

⚠️ Important: Use full absolute paths to the Python executable in your virtual environment. Replace yourname/YourName with your actual username.

Step 3: Restart Claude Desktop

After saving the config file, completely quit and restart Claude Desktop for the changes to take effect.

Step 4: Verify Connection

In Claude Desktop, you should see a hammer icon (🔨) indicating MCP tools are available. Try asking:

"Show my MyFitnessPal diary for today"

Authentication Methods

The MCP server supports four authentication methods, tried in this order:

1. Environment Variables (Legacy)

Set MFP_USERNAME and MFP_PASSWORD in your Claude Desktop config's env section. You can store them as plain text or encrypted (see below).

⚠️ Note: MyFitnessPal migrated to a NextAuth backend, so the form-POST flow this method uses no longer produces a session cookie. Credential auth only succeeds while ~/.mfp_mcp/cookies.json still holds a valid session from a previous browser login — after that, this method silently falls through to the browser cookie paths below. Prefer the Chromium auto-discovery method on macOS.

"env": {
  "MFP_USERNAME": "[email protected]",
  "MFP_PASSWORD": "your_password"
}

Encrypted Credentials (Enhanced Security)

Instead of storing plain-text credentials, you can encrypt them using Fernet symmetric encryption from the cryptography library. The server decrypts them at runtime using MFP_SECRET_KEY.

⚠️ Important: For encryption to be meaningful, MFP_SECRET_KEY must be kept outside the Claude Desktop config file. The server resolves it in this order: 1. MFP_SECRET_KEY environment variable (shell profile, not the Claude config) 2. OS keychain — service mfp-mcp, account MFP_SECRET_KEY (recommended)

Step 1 — Generate and store the key in one command:

npm install
npm run store-key

store-key generates a Fernet-compatible key, stores it in the OS keychain (mfp-mcp / MFP_SECRET_KEY), and prints the key so you can use it in Step 2. See Key Management CLI for all available flags.

Step 2 — Encrypt your credentials:

from cryptography.fernet import Fernet

key = b"abc123XYZ...=="  # your key from Step 1
f = Fernet(key)

encrypted_user = f.encrypt(b"[email protected]").decode()
encrypted_pass = f.encrypt(b"your_password").decode()

print("MFP_USERNAME:", encrypted_user)
print("MFP_PASSWORD:", encrypted_pass)

Step 3 — Add only the encrypted values to your Claude Desktop config:

"env": {
  "MFP_USERNAME": "gAAAAAB...<encrypted>",
  "MFP_PASSWORD": "gAAAAAB...<encrypted>"
}

The key stays in the keychain — it never touches the config file.

Alternative: shell profile (simpler, still outside the Claude config):

# Add to ~/.zshrc or ~/.bashrc — do NOT put this in claude_desktop_config.json
export MFP_SECRET_KEY="abc123XYZ...=="

If MFP_SECRET_KEY is not found in the environment or keychain, the server treats MFP_USERNAME and MFP_PASSWORD as plain text (backward compatible).

2. Stored Session Cookies

After successful authentication, session cookies are saved to ~/.mfp_mcp/cookies.json. These persist for 30 days, so you won't need to re-authenticate frequently.

3. Chromium Browser Auto-Discovery (macOS)

If no credentials are provided and stored cookies are absent or expired, the server scans the macOS keychain for <Browser> Safe Storage entries to find every installed Chromium-based browser, then tries each one's cookies database until it finds a valid MyFitnessPal session token.

This works out of the box with Arc, Chrome, Edge, Brave, Vivaldi, Opera, Chromium, and any other Chromium-derived browser. You only need to be logged into myfitnesspal.com in one of them.

The first successful extraction is persisted to ~/.mfp_mcp/cookies.json, so subsequent calls skip the discovery step until the session expires.

You can also force a specific browser via the refresh_browser_cookies MCP tool:

refresh_browser_cookies(browser="arc")     # or "chrome", "edge", "brave", ...
refresh_browser_cookies(browser="auto")    # scan everything (default)
refresh_browser_cookies(browser="firefox") # via browser_cookie3

4. browser_cookie3 Fallback (Legacy)

A final fallback uses browser_cookie3 to read Chrome or Firefox cookies from the default profile paths. Useful on Linux/Windows or if the macOS auto-discovery path can't access your keychain.

Security Note on Credentials

Your MyFitnessPal credentials in the Claude Desktop config are stored locally on your machine. The config file is only readable by your user account. Options to harden this further:

1. Encrypt credentials + store the key in the OS keychain

The strongest option. The ciphertext lives in the config; the key never does. See Encrypted Credentials.

2. Encrypt credentials + export the key in your shell profile

Still separates key from ciphertext, though the key is on disk.

3. Storing MFP_PASSWORD in your MCP client

Storing MFP_PASSWORD in your MCP client config puts your MyFitnessPal password in plaintext on disk, readable by anything running as your user. It is convenient — the server can re-authenticate indefinitely — but it is a real tradeoff, not a formality.

4. Use browser cookies instead (no credentials stored in config at all)

Prefer browser-cookie auth if you would rather not store the password: log into myfitnesspal.com and the server reads the session from your browser. The cost is that MFP sessions expire, so you will occasionally need to log in again.

Files this server writes to ~/.mfp_mcp/ (directory mode 0700):

Note that this server can modify your diary — adding food entries and updating goals, measurements, and water.

Usage Examples

Once configured, you can interact with your MyFitnessPal data through Claude:

Food Diary

"Show me what I ate today"
"Get my food diary for 2026-01-05"
"What meals did I log yesterday?"

Logging and Correcting Food

"Log a grilled chicken breast, 6 oz, for lunch"
"Add 2 cups of oatmeal to breakfast"

Track Weight Progress

"Show my weight history for the past 30 days"
"Log my weight as 232.5 pounds"
"What's my weight trend this month?"

Search Foods

"Search MyFitnessPal for chicken breast"
"Find nutrition info for Greek yogurt"
"Look up calories in a banana"

Check Goals vs Actual

"Compare my nutrition goals to what I actually ate today"
"Am I on track with my protein intake?"
"How many calories do I have left today?"

Exercise Log

"What exercises did I log today?"
"Show my workout from yesterday"

Nutrition Reports

"Show my calorie intake over the past week"
"What's my average protein intake this week?"
"Generate a nutrition report for January"

Key Management CLI

scripts/store-key.ts is a one-time setup tool that generates and stores MFP_SECRET_KEY in your OS keychain (macOS Keychain, Windows Credential Vault, Linux Secret Service). Node.js 18+ is required.

Prerequisites

npm install

Commands

Example output

✅ MFP_SECRET_KEY stored in OS keychain
   service : mfp-mcp
   account : MFP_SECRET_KEY
   source  : generated

Your key (use this to encrypt MFP_USERNAME / MFP_PASSWORD):

  abc123XYZ...==

Next — encrypt your credentials with Python:

  from cryptography.fernet import Fernet
  f = Fernet(b"abc123XYZ...==")
  print("MFP_USERNAME:", f.encrypt(b"[email protected]").decode())
  print("MFP_PASSWORD:", f.encrypt(b"your_password").decode())

Project Structure

myfitnesspal-mcp-python/
├── Dockerfile              # Container deployment
├── package.json            # Node tooling (store-key CLI)
├── tsconfig.json           # TypeScript config for scripts/
├── pyproject.toml          # Python package configuration
├── README.md               # This file
├── scripts/
│   └── store-key.ts        # One-time key management CLI
└── src/
    └── mfp_mcp/
        ├── __init__.py     # Package initialization
        └── server.py       # MCP server implementation

Development

Setup Development Environment

# Clone and enter directory
git clone https://github.com/YOUR_USERNAME/myfitnesspal-mcp-python.git
cd myfitnesspal-mcp-python

# Create virtual environment (Python 3.10+ required)
python3 -m venv venv
source venv/bin/activate

# Upgrade pip and install with dev dependencies
pip install --upgrade pip
pip install -e ".[dev]"

Run Tests

pytest

Code Formatting

black src/
isort src/
ruff check src/

Type Checking

mypy src/

Docker Deployment

⚠️ Note: Docker deployment requires mounting your browser's cookie database for authentication.

# Build the image
docker build -t mfp-mcp .

# Run with Chrome cookies mounted (Linux example)
docker run -it --rm \
  -v ~/.config/google-chrome:/root/.config/google-chrome:ro \
  mfp-mcp

Troubleshooting

"python: command not found" or wrong Python version

Problem: Python is not in PATH or you need to specify version.

Solutions:

1. On macOS/Linux, use python3 instead of python
2. Check your version: python3 --version (must be 3.10+)
3. If needed, install Python 3.12 via Homebrew: brew install [email protected]
4. Then create venv with: python3.12 -m venv venv

"pip install -e ." fails with "setup.py not found"

Problem: Your pip version is too old to support pyproject.toml builds.

Solution: Upgrade pip first:

pip install --upgrade pip
pip install -e .

"Failed to authenticate with MyFitnessPal"

Problem: The server can't authenticate with your credentials or read browser cookies.

Solutions:

1. Easiest (macOS): Log into myfitnesspal.com

in any Chromium-based browser (Arc, Chrome, Edge, Brave, ...). The MCP will auto-discover the session on the next call.

1. Force a refresh: Call the refresh_browser_cookies tool — auto

scans every browser, or pass a specific name (arc, chrome, edge, brave, vivaldi, opera, firefox).

1. If using credentials: Double-check your MFP_USERNAME and MFP_PASSWORD

in the config. Note that the legacy form-login flow no longer works against MFP's NextAuth backend — credentials are only useful while ~/.mfp_mcp/cookies.json still holds a valid session.

1. Try logging out and back in to MyFitnessPal in your browser.
2. Clear ~/.mfp_mcp/cookies.json and let the auto-discovery rebuild it.
3. On macOS, the auto-discovery path reads each browser's Safe Storage

password from your login keychain. On the very first run, macOS shows a dialog: "<binary> wants to use information stored in your keychain" — click Always Allow. If Claude Desktop is spawning the MCP headlessly in the background, this dialog can be easy to miss; if auto-discovery returns "no browser had a session", bring Claude Desktop to the foreground and retry so the prompt is visible. Once approved, the key is cached and the prompt won't repeat.

"No module named 'mfp_mcp'"

Problem: Package not installed or wrong Python environment.

Solutions:

1. Ensure you're using the correct Python from your virtual environment
2. Reinstall the package: pip install -e .
3. Verify the path in your Claude Desktop config points to the venv Python:

   /path/to/project/venv/bin/python  # macOS/Linux
   C:\path\to\project\venv\Scripts\python.exe  # Windows

Tools not appearing in Claude Desktop

Problem: MCP server not connecting.

Solutions:

1. Check the config file syntax (must be valid JSON - use a JSON validator)
2. Use absolute paths in the configuration (no ~ or relative paths)
3. Restart Claude Desktop completely (Cmd+Q on macOS, then relaunch)
4. Check Claude Desktop logs:

• macOS: ~/Library/Logs/Claude/
• Windows: %APPDATA%\Claude\logs\

Empty responses or no data

Problem: Authentication works but no data returned.

Solutions:

1. Verify you have data logged in MyFitnessPal for the requested date
2. Check the date format (YYYY-MM-DD)
3. Try a recent date where you know you have entries

Double parentheses in terminal prompt like "((venv) )"

Problem: VS Code/Cursor Python extension bug with venv prompt.

Solutions:

1. Update the Python extension in VS Code/Cursor
2. Or manually fix the venv activate script - change line ~70 in venv/bin/activate:

   # Change from:
   PS1="("'(venv) '") ${PS1:-}"
   # To:
   PS1="(venv) ${PS1:-}"

API Reference

mfp_get_diary

Get food diary for a specific date.

• date (optional): YYYY-MM-DD format, defaults to today
• response_format: "markdown" or "json"

mfp_search_food

Search the MyFitnessPal food database.

• query (required): Search term
• limit (optional): Max results (default 10, max 50)
• response_format: "markdown" or "json"

mfp_get_food_details

Get detailed nutrition for a food item.

• mfp_id (required): MyFitnessPal food ID from search results
• response_format: "markdown" or "json"

mfp_add_food_to_diary

Add a food item to your diary for a specific meal and date.

• mfp_id (required): MyFitnessPal food ID from search results (use mfp_search_food first)
• meal (optional): Meal name - "Breakfast", "Lunch", "Dinner", or "Snacks" (default: "Breakfast")
• date (optional): YYYY-MM-DD format (default: today)
• quantity (optional): Number of servings (default: 1.0)
• unit (optional): Unit/serving size description (e.g., "1 cup", "100g")

Example workflow:

1. Use mfp_search_food to find a food item and get its mfp_id
2. Use mfp_add_food_to_diary with the mfp_id to add it to your diary

mfp_create_custom_food

Create a private custom food in your account. Returns the new food's id, which mfp_add_food_to_diary accepts.

• description (required): Food name as it appears in MFP
• calories (required): Calories per serving
• brand_name (optional): Brand; packaged = label brand, restaurant = venue, homemade = "Generic" (default: "Generic")
• serving_amount (optional): Serving size number (default: 100)
• serving_unit (optional): Serving unit, e.g. "g", "ml", "piece" (default: "g")
• Nutrients (all optional): carbs, fiber, sugar, protein, fat, saturated_fat,

polyunsaturated_fat, monounsaturated_fat, trans_fat, cholesterol (mg), sodium (mg), potassium (mg), vitamin_a, vitamin_c, calcium, iron (last four are %DV)

• country_code (optional): Label convention (default: "NL") — see the carbs note below
• public (optional): Share publicly (default: false)
• response_format: "markdown" or "json"

`carbs` is NET carbs, and `country_code` is what makes it so. The field selects which label convention your number follows, so it changes the meaning of carbs:

Sending carbs=42, fiber=8 stores 50/42 under "NL" but 42/34 without it. Pass the number straight off the label and leave country_code matching that label's origin — do not pre-subtract fibre.

MyFitnessPal has no custom-food update endpoint. To correct a food, create the corrected version and then mfp_delete_custom_food the old one.

mfp_list_own_foods

List your own custom foods, newest first. Private custom foods do not reliably appear in mfp_search_food, so this is how to find the id of something you created earlier.

• search (optional): Substring filter on the food name
• limit (optional): Max foods to return (default: 25)
• response_format: "markdown" or "json"

mfp_delete_custom_food

Delete one of your custom foods. Destructive and not recoverable; MyFitnessPal may refuse if the food is referenced by a logged diary entry.

• food_id (required): Food id from mfp_create_custom_food or mfp_list_own_foods

mfp_remove_food_from_diary

Remove a logged entry from your diary.

• entry_id (required): Diary entry id (from mfp_get_diary)

mfp_get_measurements

Get body measurement history.

• measurement (optional): "Weight", "Body Fat", "Waist", etc.
• start_date (optional): YYYY-MM-DD (default 30 days ago)
• end_date (optional): YYYY-MM-DD (default today)
• response_format: "markdown" or "json"

mfp_set_measurement

Log a body measurement for today.

• measurement (optional): Type (default "Weight")
• value (required): Numeric value

mfp_get_exercises

Get exercise log for a date.

• date (optional): YYYY-MM-DD (default today)
• response_format: "markdown" or "json"

mfp_get_goals

Get daily nutrition goals.

• date (optional): YYYY-MM-DD (default today)
• response_format: "markdown" or "json"

mfp_set_goals

Update nutrition goals.

• calories (optional): Daily calorie goal
• protein (optional): Daily protein in grams
• carbohydrates (optional): Daily carbs in grams
• fat (optional): Daily fat in grams

mfp_get_water

Get water intake for a date.

• date (optional): YYYY-MM-DD (default today)

mfp_set_water

Log water intake for a date.

• cups (required): Number of cups of water (e.g., 2.5 for 2.5 cups). Note: MyFitnessPal uses cups as the unit (1 cup = ~237ml)
• date (optional): YYYY-MM-DD format (default: today)

mfp_get_steps

Get daily step counts synced into MyFitnessPal (Apple Health, Google Fit, connected devices). Days with no synced data return steps: null.

• start_date (optional): YYYY-MM-DD (default 6 days before end_date)
• end_date (optional): YYYY-MM-DD (default today)
• response_format: "markdown" or "json"

Maximum 31 days per call (MFP's API returns one day per request).

mfp_get_report

Get nutrition report over a date range.

• report_name (optional): "Net Calories", "Protein", "Fat", "Carbs"
• start_date (optional): YYYY-MM-DD (default 7 days ago)
• end_date (optional): YYYY-MM-DD (default today)
• response_format: "markdown" or "json"

Security & Privacy

• Encrypted Credentials: Credentials can be stored as Fernet-encrypted ciphertext in your config. MFP_SECRET_KEY is resolved at runtime from the environment variable first, then the OS keychain (mfp-mcp / MFP_SECRET_KEY). See Encrypted Credentials for setup.
• OS Keychain: Storing MFP_SECRET_KEY in the native keychain (macOS Keychain, Windows Credential Vault, Linux Secret Service) means the decryption key never touches the config file or any backup.
• Plain Credentials: If MFP_SECRET_KEY is absent from both environment and keychain, MFP_USERNAME and MFP_PASSWORD are used as-is (backward compatible).
• Session Cookies: After successful authentication, session cookies are cached in ~/.mfp_mcp/cookies.json (restricted permissions) for 30 days.
• Browser Cookies: As a fallback, the server can read your browser cookies to authenticate with MyFitnessPal.
• Local Only: The server runs locally on your machine via stdio transport. No data is sent to any third-party servers.
• No External Transmission: Your MyFitnessPal data is only transmitted between your computer and MyFitnessPal's servers (myfitnesspal.com).

License

MIT License - See LICENSE file for details.

Acknowledgments

• python-myfitnesspal - The underlying library for MyFitnessPal access
• MCP Python SDK - Model Context Protocol framework
• Anthropic - Claude and the MCP specification