home-remote-mcps
A personal remote MCP gateway that exposes Garmin Connect, Home Assistant, and YouTube as MCP servers over HTTP, allowing Claude or any MCP client to query and control them from anywhere without local credentials. It includes a web UI for managing credentials and issuing API keys, with authentication delegated to an external SSO.
README
home-remote-mcps
Personal remote MCP gateway. Exposes Garmin Connect, Home Assistant, and YouTube 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. Home Assistant and YouTube have normal REST
APIs, so the backend talks to them directly.
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-authinstance (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.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。