
Train
Recovery, HRV and training load become context before the workout starts.
Open-source MCP connectors for WHOOP, Oura, Garmin, Strava, Fitbit, Withings, Apple Health and Polar. Use them in Claude, Cursor, ChatGPT, Hermes or any MCP client. Agents never receive your tokens.
Works with every MCP-compatible client
Real-world context
Delx Wellness is not another dashboard. It turns the signals from your devices into useful context when an agent is helping with training, recovery and daily planning.

Recovery, HRV and training load become context before the workout starts.

Runs, rides and routes stay provider-tagged for cleaner agent reasoning.

Your assistant can connect today's plan with yesterday's recovery.
How it works
Each connector is a standalone npm package. Local-first by design, read-only by default, privacy modes built in.
One npx command per connector. No global install, no Docker, no servers to babysit.
Browser-based OAuth or local helper. Tokens are saved to ~/.<provider>-mcp/ with chmod 600.
Drop the snippet into Claude Desktop, Cursor, Hermes or any MCP client. Your AI now has provider-tagged health context.
Connectors
8 public connectors stay independently installable, read-only by default, and tagged agent_ready in the registry. Private labs can live beside them without exposing sensitive provider details.
Recovery score, HRV, sleep stages and day strain.
Daily readiness, sleep periods, HRV trend and SpO2.
Body Battery, training readiness, sleep, HRV and stress.
Runs, rides, swims, segments and per-second activity streams.
Activity, sleep stages, intraday heart and HRV.

Smart-scale composition, sleep stages and ECG-grade heart records.
Reads your local Apple Health export with no live HealthKit dependency.
Nightly Recharge, training load, PPI/HRV and continuous samples.
Private exercise catalog bridge for agent-built workouts.
Quick start
Pick your MCP client below, paste a snippet, then run setup for the connectors you actually use.
{
"mcpServers": {
"whoop": { "command": "npx", "args": ["-y", "whoop-mcp-unofficial"] },
"oura": { "command": "npx", "args": ["-y", "oura-mcp-unofficial"] },
"garmin": { "command": "npx", "args": ["-y", "garmin-mcp-unofficial"] },
"strava": { "command": "npx", "args": ["-y", "strava-mcp-unofficial"] },
"fitbit": { "command": "npx", "args": ["-y", "fitbit-mcp-unofficial"] },
"withings":{ "command": "npx", "args": ["-y", "withings-mcp-unofficial"] },
"apple_health": { "command": "npx", "args": ["-y", "apple-health-mcp-unofficial"] },
"polar": { "command": "npx", "args": ["-y", "polar-mcp-unofficial"] }
}
}In your AI
Stop pasting numbers from five apps. Ask once. Your agent reasons across providers in real time.
whoop_daily_summaryoura_daily_summarygarmin_daily_summaryNo — go light. Your signals are mixed-to-yellow today.
Suggestion: zone-2 cardio 30–40 min, mobility, sleep priority. Re-evaluate tomorrow with whoop_daily_performance_coach.
Not medical advice.
Privacy & security
Health data is sensitive — these connectors are built to treat it that way. There is no hosted layer in this registry. There is no place to leak your data, because there is no place to put it.

OAuth tokens are saved in ~/.<provider>-mcp/ with chmod 600. Nothing is uploaded to a hub. No third-party vault.
No upload scopes are requested. The MCP client never receives access or refresh tokens. Tools never echo secrets.
Every connector ships *_privacy_audit and *_connection_status tools so agents can verify posture without calling the upstream API.
summary / structured / raw — defaults are conservative. Raw upstream payloads and GPS coordinates are explicit opt-in.
FAQ
Model Context Protocol — an open standard for connecting AI assistants to external tools and data. Once a server speaks MCP, compatible clients such as Claude Desktop, Cursor, ChatGPT Desktop, Hermes and custom agents can use its tools.
Health data is sensitive. Hosted hubs concentrate secrets and become attractive targets. The connectors here run on your machine, store tokens locally with chmod 600, and never share them with the MCP client. A hosted layer may exist later, but it will be a separate, opt-in product.
Yes, but you probably should start with the providers you already use. The packages are independent and coexist cleanly. When two providers cover the same signal, each tool returns provider-tagged data so your agent can reconcile them.
No. Every connector is unofficial and not affiliated with the device makers. They use public APIs where available; Garmin uses the user's own Garmin Connect login locally, and Apple Health reads a manual export.zip rather than live HealthKit.
No. These connectors expose user-authorized data for personal AI workflows. They are not medical devices and do not provide diagnosis, treatment or emergency monitoring. Always consult qualified professionals for medical concerns.
Yes — the connector quality standard documents the bar (registry_ready / agent_ready). Open an issue or PR on the delx-wellness repo with the provider you want to add and the data you'd expose.
Eight open-source connectors. One local-first setup. Your AI can reason over sleep, recovery, activity and training context.