Não uses o MCP da Strava para dados do Garmin — usa isto em vez disso

Day 105: Don't Use the Strava MCP If You Want the Full Picture

Strava launched an official MCP connector for Claude and ChatGPT, and it's a genuinely good tool — read-only, scoped to your account, built by Strava itself. But if you're wearing a Garmin, it's the wrong connector to reach for.

What the Strava MCP covers

Strava's connector focuses on activity history, fitness trends, readiness, goal planning, and gear — everything tied to a logged workout. What it doesn't expose is the recovery side: sleep stages, overnight HRV, stress, or Body Battery. That data never leaves your Garmin account in the first place.

What the Garmin MCP adds on top

A Garmin MCP server gives Claude your workout data too — activities, training load, VO2 max, training status — but layers your full health picture on top:

• Sleep stages, sleep score, and sleep consistency
• Overnight HRV and HRV Status trend
• Stress levels and Body Battery charge/drain patterns
• Resting heart rate and training readiness

That's the difference between Claude seeing what you did and Claude seeing why you're able (or unable) to do it. A hard week of training means something very different paired with a suppressed HRV trend and poor sleep than it does paired with strong recovery numbers.

Setting it up

This uses the free, open-source garmin_mcp project. No coding experience needed — every step is copy-paste.

Step 1: Open a terminal

On Mac: Press Cmd + Space, type Terminal, and press Enter.

On Windows: Press the Windows key, type PowerShell, and press Enter.

Step 2: Install uv (a small required tool)

The Garmin MCP server runs through uv, a free tool that downloads and runs it for you — including Python itself, so you don't need to install Python separately. Paste the command for your system into the terminal and press Enter:

Mac:

curl -LsSf https://astral.sh/uv/install.sh | sh

Windows (PowerShell):

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

Text will scroll by — that's normal. When it stops and shows a new cursor, it's done. You only ever need to do this once on this computer.

Step 3: Log in to Garmin, from the terminal

Still in the same terminal window, paste this and press Enter (the first run may take a minute, since it also downloads a small Python runtime automatically):

uvx --python 3.12 --from git+https://github.com/Taxuspt/garmin_mcp garmin-mcp-auth

It will ask for your Garmin email, password, and — if you have it enabled — your MFA code. Type each one and press Enter. This saves a secure login token on your computer, valid for about six months, so Claude will never have to ask you to log in again.

Step 4: Download the Garmin MCP extension

Go to the project's Releases page and download the latest garmin-mcp.dxt file, listed under "Assets".

Step 5: Install it in Claude Desktop

Open the Claude desktop app (download and install it first if you haven't already), then drag the garmin-mcp.dxt file you just downloaded straight onto the Claude Desktop window. You can also double-click the file, or go to Settings → Extensions → Install Extension and pick it manually.

Claude shows a short configuration screen for the extension, with optional fields like token path, email, and password. Because you already logged in during Step 3, leave every field empty and just confirm — Claude uses your saved login automatically.

Step 6: Restart Claude Desktop

Fully quit and reopen Claude Desktop, then start a new conversation. Your Garmin data is now available to Claude directly, with no export and no copy-pasting.

If something goes wrong: if Claude says it can't find uvx, type which uvx (Mac) into the terminal, copy the path it prints, and paste that full path into the extension's "command" field instead of uvx. Your login token lasts about six months — if Claude suddenly stops seeing your data, just repeat Step 3.

Put it to work: build a real training plan

Once it's connected, this is the prompt to use. State your goal first, then let Claude pull your actual data before it plans anything:

[explain your goal here, for example: run 10 miles in 6 months]

Before building a plan, pull my last 14 days of training and health data from Garmin:
- All activities: type, distance, duration, pace, average/max HR, elevation, perceived effort if logged
- Training load and Training Status (productive, maintaining, overreaching, etc.)
- VO2max trend
- Resting HR and HRV trend
- Sleep score average and consistency
- Body Battery and stress trends
- Any current injuries, pain, or recovery flags if mentioned in activity notes

Use this data to establish my current baseline fitness level, honestly — don't assume a generic starting point. Then build a training plan to reach the stated goal by the target date.

The plan should:
- Start from where my data actually shows I am now (weekly volume, pace, long-run distance, recovery capacity), not an idealized starting point
- Progress volume/intensity gradually, respecting a sensible weekly increase (~10% or less) and account for my current Training Status and recovery trends — back off progression if HRV/sleep/Body Battery show I'm already under strain
- Include a weekly structure (easy runs, one quality session, long run, rest/cross-train days) appropriate for the goal
- Flag any specific risk areas based on my data (e.g. low resting heart rate variability, poor sleep consistency, high training load already) and how the plan accounts for them
- Break the plan into phases (e.g. base building, build, peak, taper if relevant) with approximate weekly mileage/duration targets per phase
- Note 2-3 checkpoints where I should reassess (e.g. week 4, week 10) based on how my Garmin metrics are trending, not just the calendar

Output as a clean, well-formatted document (Word doc): summary of current baseline, the phased plan with a week-by-week or phase-by-phase table, key risks flagged, and reassessment checkpoints. Keep language direct and practical, no fluff.

Because Claude reads your Garmin account directly through the MCP connection, there's no export step and no stale data — every time you run this, it's working from your latest numbers.