Skip to main content
Glama

home-remote-mcps

Personal remote MCP gateway. Exposes Garmin Connect, Home Assistant, YouTube, personal health data, and Docker container logs (shipped to MinIO by Vector) as Model Context Protocol servers reachable over HTTP, so Claude (or any MCP client) can query and control them from anywhere — no local server, no local credentials.

Authentication for the web UI is delegated to an external SSO (home-auth, OAuth2). MCP clients authenticate separately with a per-user API key issued from that UI.

Why a sidecar for Garmin

Garmin Connect has no official public API and no actively maintained Node client. The garmin-connector service is a small internal-only Python sidecar wrapping the garminconnect library, which handles Garmin's login flow (including MFA) and token refresh. The NestJS backend talks to it over the Docker network with a shared secret; it is never exposed publicly. Every other integration — Home Assistant, YouTube, personal health, and logs — has a normal REST (or S3-compatible) API, so the backend talks to them directly with no sidecar.

Related MCP server: FitnessMCP

Project layout

backend/            NestJS API — REST (auth, credentials, API keys) + MCP servers, DDD-ish layers
frontend/            React SPA — login, manage stored credentials, issue MCP API keys
garmin-connector/    FastAPI sidecar — Garmin Connect login/session via `garminconnect`
deploy/              docker-compose + Caddy config for production

See CLAUDE.md for the detailed architecture (layering, MCP request path, per-integration connector design, auth model) and ai-docs/ for deeper reference material. Human-oriented guides live in docs/.

Requirements

  • Node.js >= 20 (see .nvmrc)

  • Python >= 3.11 for garmin-connector

  • A running home-auth instance (OAuth2 client registered for this app) for browser login

  • Docker, for production deploys (deploy/docker-compose.yml)

Local development

npm install                    # installs backend + frontend workspaces

cp backend/.env.example backend/.env          # fill in home-auth + secrets, see comments in the file
cp garmin-connector/.env.example garmin-connector/.env

npm run dev:backend            # http://localhost:3000 (API under /api, Swagger under /api/docs)
npm run dev:frontend           # http://localhost:5173, proxies API calls to the backend

cd garmin-connector && pip install -r requirements.txt && uvicorn app.main:app --reload --port 8000

GARMIN_CONNECTOR_SECRET must match between backend/.env and garmin-connector/.env.

Build, test, lint

npm run build        # frontend, then backend (bundles frontend/dist into the backend build)
npm run test          # backend (jest) then frontend (vitest)
npm run lint           # eslint on both workspaces
npm run format          # prettier --write, repo-wide

Husky hooks (pre-commit, pre-push, commit-msg) run lint-staged and commitlint (Conventional Commits) automatically.

Deployment

Dockerfile at the repo root multi-stage builds frontend → backend → a runtime image that bundles Litestream for continuous SQLite replication to a MinIO (S3-compatible) bucket, restoring on boot if MINIO_BUCKET is set. garmin-connector/Dockerfile builds the sidecar image separately. deploy/docker-compose.yml runs both as sibling services on an internal Docker network — the sidecar has no published port, reachable only from the backend at http://garmin-connector:8000. deploy/Caddyfile is the reverse-proxy config for the public service.

Required environment variables are documented inline in backend/.env.example and garmin-connector/.env.example.

Related MCP Connectors

  • Zero-setup MCP gateway securely connecting AI to your tools with authentication and workflows

  • Remote streamable-HTTP MCP server running on a single Cloudflare Worker. Your assistant gets live Airbnb, Amazon, Booking.com, Google Flights, Maps and Reddit data, social search on X, Instagram and TikTok, the Meta Ad Library, and image/video generation without any keys. Connect your own accounts to let it send WhatsApp or Telegram messages, work an IMAP inbox, manage Meta Ads campaigns and publish to X and LinkedIn. OAuth 2.1 with PKCE; stored credentials are AES-256-GCM encrypted.

  • An authenticated remote MCP server for user-owned devices and one-shot capability invocation.

  • The Remote MCP server acts as a standardized bridge between LLM applications (like Claude, ChatGPT, and Cursor) and external services, enabling AI agents to access external tools and resources. Its primary capability is providing a centralized search tool to discover other MCP servers and their respective tools. Unlike local implementations, it runs remotely with OAuth authentication and permission controls for security.

Related MCP Servers

  • F
    license
    Not graded
    quality
    D
    maintenance
    A unified data gateway MCP server that enables Claude to interact with multiple external REST and GraphQL APIs through OAuth 2.0 authentication.
    -
  • A
    license
    Not graded
    quality
    D
    maintenance
    A unified MCP server that connects AI assistants to multiple fitness services (Hevy, Strava, Cronometer, Intervals.icu) through a single secure endpoint, enabling workout, nutrition, and activity data access.
    MIT