Skip to main content
Glama
deenesjakoruzh
nnishad

health-relay

by nnishad
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Hybrid

health-relay

Self-hosted bridge between your Samsung Watch → Sync2Ra1 phone app and your Hermes AI agent. The app pushes health readings here over your LAN; the data is stored locally (SQLite) and exposed via REST and MCP so your agent can query it.

Galaxy Watch ─▶ Samsung Health ─▶ Sync2Ra1 (phone) ──push──▶ health-relay ──MCP/REST──▶ Hermes

All data stays on your network. Nothing is sent anywhere else.

Quick start (Docker or Podman)

echo "HR_AUTH_TOKEN=<choose-a-secret>" > .env
docker compose up -d          # or: podman compose up -d
curl http://localhost:8790/healthz     # -> {"ok":true}

Data persists in ./data/ledger.db (SQLite, WAL mode).

Run without Docker

python3 -m venv .venv && .venv/bin/pip install -r requirements.txt
HR_AUTH_TOKEN=<secret> ./run.sh

Related MCP server: Yummy Health MCP

Point the phone app at it

In Sync2Ra1 → Data tab → Relay (Hermes bridge):

Field

Value

Relay URL

http://<this-machine-ip>:8790

Token

same secret as HR_AUTH_TOKEN

Tap Save → Sync now. Readings arrive within seconds.

Connectivity

phone (Sync2Ra1) ──HTTP :8790──▶ health-relay ◀──MCP stdio / HTTP── Hermes
        same LAN · or Tailscale for anywhere-access

Rule

Detail

Direction

Phone → relay (outbound from phone; nothing connects into the phone)

Port

TCP 8790 inbound to the machine running the relay — open it in that host's firewall

Protocol

Plain HTTP + Authorization: Bearer <token>. Keep it LAN-only, or front with a reverse proxy (Caddy/nginx) for TLS

Anywhere-access

Install Tailscale on the box and phone; point the app at the Tailscale IP (http://100.x.y.z:8790) — encrypted, works over mobile data, no ports exposed to the internet

Cadence

App auto-syncs hourly + on app launch + manual Sync now

Offline

Phone buffers everything in local SQLite; on reconnect it pushes the full backlog (oldest first) and re-syncs the last 48 h so Samsung's revisions propagate

Duplicates

Impossible to double-count — every reading has a unique key; replays are ignored server-side

Hermes location

Same machine (MCP runs as a local subprocess reading the SQLite file) or remote (use the REST endpoints instead of MCP)

REST API (bearer auth)

Endpoint

Purpose

GET /healthz

liveness (no auth)

POST /api/v1/health/samples

batch ingest — idempotent, replays are no-ops

GET /api/v1/health/samples?type=&from_ms=&to_ms=&limit=&offset=

raw readings

GET /api/v1/health/daily-summary?date=YYYY-MM-DD&tz=Europe/London

per-metric min/max/avg/count

GET /api/v1/health/trends?type=X&days=N&tz=Europe/London

daily rollups

GET /api/v1/health/status

totals per type

curl -H "Authorization: Bearer <secret>" \
  "http://localhost:8790/api/v1/health/daily-summary?date=$(date +%F)&tz=Europe/London"

Plug into Hermes (MCP)

The MCP server reads the same SQLite file — run it from this repo:

python3 -m venv .venv-mcp
.venv-mcp/bin/pip install fastmcp

Add to your agent's MCP config:

{
  "mcpServers": {
    "health-relay": {
      "command": "<repo-path>/.venv-mcp/bin/python",
      "args": ["<repo-path>/mcp_server.py"],
      "env": { "HR_DATA_DIR": "<repo-path>/data" }
    }
  }
}

Tools your agent gets

Tool

Answers questions like

list_streams()

"what health data do you have?"

get_samples(type, from_ms, to_ms)

"show my heart rate this morning"

daily_summary(date, tz)

"how active was I yesterday?"

sleep_breakdown(nights)

"how did I sleep this week?"

Day boundaries respect your timezone (tz="Europe/London" etc., DST handled).

Recorded streams

Heart rate · SpO₂ · skin temperature · respiration · resting HR · VO₂max · sleep duration/score/stages · steps & floors (hourly + daily) · workouts · activity calories/time/distance · body composition (BIA) · energy score · blood pressure · glucose · hydration · meals.

Configuration

Env var

Default

HR_AUTH_TOKEN

(empty = auth off)

set this in production

HR_PORT

8790

HR_HOST

0.0.0.0

keep on trusted LAN

HR_DATA_DIR

~/health-relay-data

SQLite lives here

Tests

.venv/bin/python -m pytest tests/ -q

This server cannot be deployed

One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

Maintenance

ActivityMaintained
ResponsivenessNo issues

Resources

Related MCP Connectors

Related MCP Servers

  • Yummy Health MCP

    yummy520730
    F
    license
    Not graded
    quality
    B
    maintenance
    Private, read-only health data bridge that reads from Android Health Connect and exposes heart rate, steps, sleep, and other health metrics to ChatGPT via MCP tools.
    -

  • garmin-owl

    xichen-de
    A
    license
    B
    quality
    A
    maintenance
    Enables local, read-only access to Garmin Connect health and training data through MCP, letting assistants summarize recovery, sleep, activities, and more without uploading your data.
    19
    MIT