google-health-mcp
Read-only MCP server exposing daily health facts, 7/28-day baselines, and deterministic daily pulse from Google Health API data.
README
Google Health Daily Pulse
A self-hosted health service for Google Health API v4. It synchronizes Fitbit, Pixel Watch, and compatible health data, preserves timestamped records locally, calculates reproducible 7-day and 28-day comparisons, and exposes both summaries and full-fidelity data to Hermes Agent through a read-only MCP server.
You can do anything you want with this repo. Codex was used to code it. Use at your own risk. For Mac users, there should not be much of a difference. I will test it on a Mac later and update the repository if there are any issues.
Architecture
Google Health API
→ OAuth 2.0 + encrypted tokens
→ FastAPI / healthctl
→ SQLCipher-encrypted database + encrypted sensitive payloads
→ deterministic baselines
→ local read-only MCP server
→ Hermes (optional)
→ Discord
The health core works without Hermes. If Hermes, the subscription, or a rate limit is unavailable, the service can produce a deterministic English template message.
Quick start
Requirement: Python 3.11 or newer.
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -e ".[dev]"
Copy-Item .env.example .env
healthctl seed-demo
healthctl daily-pulse --no-hermes
Alternatively, healthctl setup creates .env interactively without displaying the
client secret. OAuth client credentials are stored in data/credentials.enc, not
.env.
Existing installations should run this once after upgrading:
healthctl migrate-security
This converts an existing plaintext SQLite database to SQLCipher, moves Google OAuth
client credentials out of .env, protects the master key with Windows DPAPI, and
restricts local file permissions.
For real data:
- Configure Google Cloud.
- Run
healthctl serve. - In a second terminal, run
healthctl authorize. - Run
healthctl sync. - Run
healthctl daily-pulse --no-hermes.
See docs/windows-setup.md for the complete Windows, Hermes, Codex, Discord, MCP, and scheduler setup.
What needs to stay running?
healthctl serve and the MCP server are separate:
healthctl servestarts the local FastAPI service on127.0.0.1:8765. It is needed for the Google OAuth callback and only if you want to use the REST endpoints.- The local MCP server uses stdio. Hermes starts
google-health-mcpautomatically when it loads the configuredgoogle-healthMCP connection. - You do not need to keep
healthctl serverunning for MCP access, synchronization through the CLI, or daily-pulse generation. - To use health tools from Discord continuously, the Hermes gateway must be running.
- To refresh the local database, run
healthctl syncmanually or schedulehealthctl run-daily.
REST API
The service binds only to 127.0.0.1:8765 by default.
healthctl serve
| Method | Path | Purpose |
|---|---|---|
GET |
/healthz |
Configuration and sync status without secrets |
GET |
/oauth/google/start |
Start Google OAuth |
GET |
/oauth/google/callback |
Receive the OAuth callback |
DELETE |
/oauth/google |
Revoke Google access |
POST |
/v1/sync |
Synchronize a date range |
POST |
/v1/daily-pulse |
Generate facts and a message |
Examples:
Invoke-RestMethod -Method Post `
-Uri http://localhost:8765/v1/sync `
-ContentType application/json `
-Body '{"start_date":"2026-05-15","end_date":"2026-06-20"}'
Invoke-RestMethod -Method Post `
-Uri http://localhost:8765/v1/daily-pulse `
-ContentType application/json `
-Body '{"use_hermes":true,"send":false}'
Uvicorn access logs are disabled by the CLI server so OAuth codes do not appear in query-string logs.
Data model
The database stores two complementary layers.
Normalized daily values:
- Sleep duration, sleep period, awake time, awakenings, and efficiency
- Light, deep, and REM sleep when available
- Steps
- Light, moderate, and vigorous activity minutes
- Exercise count, duration, and types
- Daily resting heart rate
- Daily HRV/RMSSD
Full-fidelity records preserve the complete structured Google payload and timestamps for every supported stream under the configured activity, health-metrics, and sleep scopes. This includes:
- Complete sleep sessions, stage segments, and out-of-bed segments
- Intraday heart rate, HRV, oxygen saturation, and respiratory measurements
- Heart-rate zones, activity intervals, sedentary periods, and exercise details
- Energy, distance, floors, VO2 max, body measurements, and temperature records
- Daily rollups for total calories and calories in heart-rate zones
The complete SQLite database, including normalized metrics, timestamps, indexes,
OAuth states, and tokens, is encrypted with SQLCipher. Sensitive payloads and OAuth
tokens retain an additional Fernet encryption layer. Google OAuth client credentials
are stored in an encrypted credential file rather than .env.
On Windows, the local master key at data/token.key is wrapped with DPAPI for the
current Windows account. The data directory, database, key, credential store, and
.env use ACLs restricted to that account and SYSTEM. Secrets must be decrypted in
application memory while in use; no application can protect them from malware already
running as the same signed-in user.
The default daily sync uses 35 days for compact baseline metrics and two recent days
for full-resolution records. This avoids re-downloading hundreds of thousands of
one-second heart-rate samples every morning. Detailed records accumulate locally as
the scheduled sync advances. Configure HEALTH_DETAILED_SYNC_LOOKBACK_DAYS from 1 to
90 for a larger initial backfill.
CLI
healthctl setup
healthctl migrate-security
healthctl serve
healthctl authorize
healthctl sync
healthctl facts [--date YYYY-MM-DD]
healthctl daily-pulse [--send] [--no-hermes]
healthctl run-daily [--send]
healthctl seed-demo
healthctl doctor
Local MCP server
The project includes a read-only stdio MCP server for Hermes and other local MCP clients:
google-health-mcp --project-dir C:\path\to\Google-Health
It exposes:
- Calculated daily facts and 7-day/28-day baselines
- The deterministic English daily pulse
- Selected normalized history for up to 90 days
- A catalog of all detailed data types and their local coverage
- Full-fidelity records for a date range with bounded pagination
- Exact timestamp-window queries for calendar and workout correlation
- Authorization, sync, and local data-coverage status
It does not expose OAuth tokens, Google credentials, synchronization, revocation, or database-write operations. Detailed health payloads are intentionally available to the local agent, but each call is date bounded, limited to at most 500 records, and pageable.
Hermes normally starts this executable automatically. You should not run it manually unless you are testing an MCP client.
Hermes skill
The repository includes skills/google-health/SKILL.md. It teaches Hermes how to choose the correct health tool, correlate exact windows with calendar events, handle timezones and personal baselines, and use memory without copying raw health data into it.
Install it into the active Hermes profile:
$SkillsRoot = if (Test-Path "$env:LOCALAPPDATA\hermes\skills") {
"$env:LOCALAPPDATA\hermes\skills"
} else {
"$HOME\.hermes\skills"
}
$Target = Join-Path $SkillsRoot "personal\google-health"
New-Item -ItemType Directory -Force $Target | Out-Null
Copy-Item .\skills\google-health\SKILL.md $Target\SKILL.md -Force
Start a new Hermes session after installing or changing the skill.
Security and product boundaries
- Never put secrets in issues, chat messages, or Git.
- Only read-only Google Health scopes are requested.
- The complete health database is encrypted at rest with SQLCipher.
- OAuth client credentials are encrypted outside
.env; tokens are encrypted inside the SQLCipher database with an additional Fernet layer. - Hermes Discord access is restricted through
DISCORD_ALLOWED_USERS. - The FastAPI service binds to localhost by default and provides no public multi-user authentication.
- Health guidance describes trends; it does not provide diagnoses or treatment.
- A centrally hosted service requires a different security architecture, privacy review, Google OAuth verification, and recurring restricted-scope security review.
Tests
pytest
ruff check .
Official references
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.