garmin-connect

Garmin Connect data access (Cloudflare-aware, June 2026)

This skill lets an agent retrieve a user's own Garmin Connect data and build automations on it. It encodes the currently working authentication recipe.

When to use

The user wants their Garmin data, or is stuck authenticating to Garmin from Python (429 / 401 / "Verify you are human" / MFA loops). Also for building daily sync, dashboards, or AI coaching on Garmin data.

Why the usual approaches fail (and what works)

Since March 2026 Garmin's SSO (sso.garmin.com) is behind Cloudflare. Plain HTTP clients and the newer native-auth garminconnect/garth get blocked or stuck on a "Verify you are human" challenge.

Working recipe: the older garth-based garminconnect==0.2.38 with a cloudscraper session injected into garth. cloudscraper transparently solves Cloudflare's JS challenge.

Prerequisite (tell the user)

Garmin Two-Step Verification must be OFF (Account → Security Center → Two-Step Verification → E-mail and SMS both Off). garminconnect has no 2FA flow. After the token is cached the user may re-enable 2FA; the token keeps working.

Step 1 — install

python3 -m venv venv
venv/bin/python -m pip install "garminconnect==0.2.38" cloudscraper python-dotenv

Step 2 — authenticate once (token lasts ~1 year)

The KEY line is injecting cloudscraper into client.garth.sess:

import cloudscraper
from garminconnect import Garmin

client = Garmin(email, password)          # 2FA must be OFF
client.garth.sess = cloudscraper.create_scraper()   # ← bypass Cloudflare
client.login()
client.garth.dump("~/.garminconnect")     # cache token (~1 year)

Run this interactively the first time (prompt the user for the password with getpass; never log it). Token is saved to ~/.garminconnect.

Step 3 — fetch data (token only, no login)

Re-inject cloudscraper for API calls too (connectapi.garmin.com is also behind Cloudflare):

import cloudscraper
from garminconnect import Garmin
c = Garmin()
c.garth.sess = cloudscraper.create_scraper()
c.login("~/.garminconnect")               # loads cached token

c.get_stats("2026-06-19")                 # steps, calories, restingHeartRate, bodyBatteryMostRecentValue, averageStressLevel
c.get_sleep_data("2026-06-19")            # dailySleepDTO: sleepTimeSeconds, sleepScores.overall.value
c.get_training_readiness("2026-06-19")    # [{score, level}]
c.get_activities(0, 10)                   # activities: distance, duration, averageSpeed, averageHR, averageRunningCadenceInStepsPerMinute

Pace from averageSpeed (m/s): (1000/speed)/60 → min/km.