Garmin in Claude. Sixty seconds, no install.
Claude.ai already supports custom connectors. Add the stats.training link in Settings → Connectors → Add custom connector, give it a name, and Claude can read your live Garmin data in every chat from that point on.
Three clicks in Claude.
- Open claude.ai › Settings › Connectors.Available on every Claude tier. Free gets one custom connector; Pro and Max are unlimited.
- Click "Add custom connector".Some accounts label this "Add MCP server". Same button.
- Paste your stats.training link, save.Claude lists out the available tools (sleep, HRV, training, runs, races, body) within a few seconds. Done.
Long context + multi-step reasoning.
Claude is good at pulling several signals together and reasoning across them. Garmin is the deepest single source of training and recovery data. Putting them in the same conversation is more than the sum of the parts.
What you get with this pairing specifically: Claude can call multiple tools in one turn (HRV + sleep + yesterday's load + body battery), then write a layered answer that crosses signals. Sample turn: Claude reads your overnight HRV, notices it's down 9% from baseline, checks the previous day's training load (high), looks at sleep score (decent at 78), and concludes that today's workout should be Z2 instead of intervals. One question, one answer. With shorter-context models you'd need to lead it through each signal manually.
First conversations.
More worked examples on /use-cases.
Claude-specific questions.
Does this work with Claude Free?
Yes. Free Claude lets you add one custom connector. Add stats.training and you've used it. Pro and Max have no limit.
Claude desktop, Claude Code, or Claude.ai web?
Anywhere Claude exposes the connectors UI. claude.ai web is the easiest place to add it. Claude Code (the CLI) reads MCP servers from a config file; you can add ours there too if you want to use it during development. Same private link, same tools.
Does Claude actually call multiple tools in one turn?
Yes. Claude routinely chains 3-5 tool calls per turn for analytical questions. We've watched it call get_hrv, get_sleep, get_body_battery, and get_activities in a single response while answering "should I run today?" That's the design intent of the pairing.
What about Sonnet vs. Opus?
Both work. Opus is better at multi-signal reasoning (planning a training block, cross-correlating sleep with load over weeks). Sonnet is plenty for "review my week" or "should I run today?" The connector is identical.