GlassCloud
A cloud relay server that bridges the GlassBridge smart glasses app with Google services like Gmail and Calendar via the Model Context Protocol. It provides secure OAuth authentication and WebSocket communication to enable voice-controlled tool execution on mobile devices.
README
GlassCloud
MCP Relay Server for GlassBridge - a cloud service that bridges the GlassBridge Android app with Google services and third-party tools via the Model Context Protocol (MCP).
Purpose
GlassCloud solves a fundamental challenge in mobile AI assistants: how do you give a voice assistant on smart glasses access to your personal data (email, calendar) securely?
The answer is a cloud relay that:
- Authenticates users via Google OAuth on a web browser
- Links devices via QR code scanning (no typing passwords on glasses)
- Proxies tool calls from the Android app to Google APIs
- Manages OAuth tokens securely with encryption at rest
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│ Smart Glasses │────▶│ GlassCloud │────▶│ Google APIs │
│ + Android App │ WS │ (This Server) │ │ Gmail/Calendar │
└─────────────────┘ └─────────────────┘ └─────────────────┘
│
▼
┌─────────────────┐
│ Web Console │
│ (OAuth + QR) │
└─────────────────┘
Key Features
- WebSocket Relay - Real-time bidirectional communication with Android devices
- Google OAuth - Secure authentication without exposing credentials to the mobile app
- QR Code Linking - Scan-to-link flow for easy device pairing
- MCP Tool Execution - Gmail and Calendar tools with automatic token refresh
- Voice-First Design - Progress messages for immediate audio feedback during tool execution
Quick Start
# Install dependencies
npm install
# Copy and configure environment
cp .env.example .env
# Edit .env with your settings (see Configuration below)
# Development (auto-reload)
npm run dev
# Production
npm run build
npm start
Then open: http://localhost:3000/console
Configuration
Required Environment Variables
# Security - MUST be unique random values (32+ chars)
# Generate with: node -e "console.log(require('crypto').randomBytes(32).toString('base64'))"
JWT_SECRET=your-random-secret-here
ENCRYPTION_KEY=your-random-key-here
# Google OAuth (optional for dev, required for production)
# Create at: https://console.cloud.google.com/apis/credentials
GOOGLE_CLIENT_ID=xxx.apps.googleusercontent.com
GOOGLE_CLIENT_SECRET=xxx
GOOGLE_CALLBACK_URL=http://localhost:3000/auth/google/callback
Optional Settings
PORT=3000 # Server port
NODE_ENV=development # development | production
LOG_LEVEL=debug # trace | debug | info | warn | error
DATABASE_PATH=./data/glasscloud.db
CORS_ORIGINS=http://localhost:3000
Architecture
System Components
┌─────────────────────────────────────────────────────────────────┐
│ GlassCloud Server │
├─────────────────────────────────────────────────────────────────┤
│ │
│ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │
│ │ WebSocket │ │ REST API │ │ MCP Proxy │ │
│ │ Server │ │ (Express) │ │ Manager │ │
│ │ │ │ │ │ │ │
│ │ - Device │ │ - OAuth │ │ - Gmail │ │
│ │ connections│ │ - QR codes │ │ - Calendar │ │
│ │ - Tool │ │ - Devices │ │ - Token │ │
│ │ routing │ │ - Console │ │ refresh │ │
│ └──────┬──────┘ └──────┬──────┘ └──────┬──────┘ │
│ │ │ │ │
│ └─────────────────┼─────────────────┘ │
│ │ │
│ ┌────────┴────────┐ │
│ │ SQLite + WAL │ │
│ │ │ │
│ │ - Users │ │
│ │ - Devices │ │
│ │ - OAuth tokens │ │
│ │ - Link tokens │ │
│ └─────────────────┘ │
│ │
└─────────────────────────────────────────────────────────────────┘
Directory Structure
src/
├── index.ts # Entry point, server startup
├── config/
│ ├── env.ts # Zod environment validation
│ ├── mcp-services.ts # Built-in service definitions
│ └── index.ts
├── server/
│ ├── express.ts # Express app setup (CORS, helmet, rate limiting)
│ └── websocket.ts # WebSocket server with zombie cleanup
├── routes/
│ ├── auth.ts # Google OAuth flow
│ ├── console.ts # Web console UI
│ ├── devices.ts # Device management API
│ ├── health.ts # Health check endpoint
│ ├── link.ts # QR code token generation
│ └── mcp.ts # MCP services API
├── websocket/
│ ├── handler.ts # Message routing with progress feedback
│ ├── protocol.ts # Message type definitions
│ └── connection.ts # Connection tracking
├── services/
│ ├── auth.service.ts # OAuth + token refresh mutex
│ ├── device.service.ts # Device CRUD operations
│ ├── link.service.ts # QR code token handling
│ └── mcp-proxy.service.ts # Tool execution + input coercion
├── mcp/
│ ├── gmail.ts # Gmail API integration
│ ├── calendar.ts # Calendar API integration
│ └── registry.ts
├── db/
│ ├── index.ts # SQLite connection + WAL mode
│ └── schema.ts # Table definitions
├── utils/
│ ├── logger.ts # Pino structured logging
│ ├── crypto.ts # AES-256-GCM encryption
│ └── cache.ts # LRU cache for tool results
└── types/
├── api.ts # REST API types
├── mcp.ts # MCP types
└── websocket.ts # WebSocket message types
Design Decisions
Why a Cloud Relay?
-
OAuth Security - Google OAuth requires a web browser redirect flow. Smart glasses can't do this, but they can scan a QR code.
-
Token Management - OAuth tokens must be refreshed periodically. Doing this on-device means storing refresh tokens on the phone. The relay handles this centrally.
-
Connection Stability - Mobile connections are flaky. The relay maintains persistent connections to Google APIs while tolerating device disconnects.
SQLite with WAL Mode
We use SQLite instead of PostgreSQL for simplicity:
- Zero configuration - No separate database server
- Faster for single-instance - No network latency
- WAL mode - Enables concurrent reads during writes
db.pragma('journal_mode = WAL');
db.pragma('synchronous = NORMAL');
Token Refresh Mutex
When a user asks "Check my email and add a meeting", the LLM might fire two tool calls simultaneously. Without protection, both could try to refresh an expired OAuth token, causing one to fail.
Solution: A promise-based mutex that makes concurrent refresh requests wait for the first one.
const refreshPromises = new Map<string, Promise<Token>>();
async function getValidToken(userId: string) {
// If refresh in progress, wait for it
const existing = refreshPromises.get(userId);
if (existing) return existing;
// Start new refresh
const promise = refreshToken(userId);
refreshPromises.set(userId, promise);
// ...
}
Voice-First UX
Tool execution can take 2-5 seconds. In a voice app, silence feels broken.
Solution: Send tool_progress immediately when execution starts:
{"type": "tool_progress", "status": "executing", "message": "Checking your emails..."}
The Android app can play a "thinking" sound while waiting.
Content Truncation
Large emails (10MB with attachments) would crash the Android JSON parser.
Solution: Truncate to 10KB and tell the LLM:
[...Email truncated due to size. Full content not available...]
This prevents the LLM from hallucinating the rest of the email.
Input Coercion
LLMs often send "10" (string) when the schema expects 10 (number).
Solution: Use Zod with coercion:
z.coerce.number().int().min(1).max(50)
// Accepts both 10 and "10"
WebSocket Protocol
Connection
ws://localhost:3000/ws?deviceId=UNIQUE_DEVICE_ID
Client → Server Messages
// Execute a tool
{ "type": "tool_execute", "requestId": "uuid", "serverId": "gmail",
"toolName": "gmail.get_unread", "arguments": { "maxResults": 10 } }
// List available servers
{ "type": "get_servers", "requestId": "uuid" }
// Link device to user
{ "type": "link_device", "requestId": "uuid", "linkToken": "from-qr-code", "deviceId": "..." }
// Get user account info
{ "type": "get_user_account", "requestId": "uuid", "deviceId": "..." }
Server → Client Messages
// Tool execution started (for voice feedback)
{ "type": "tool_progress", "requestId": "uuid", "status": "executing",
"message": "Checking your emails..." }
// Tool result
{ "type": "tool_result", "requestId": "uuid",
"result": { "success": true, "isError": false, "content": "You have 3 unread emails..." } }
// Available servers
{ "type": "servers_list", "requestId": "uuid", "servers": [...] }
// Error
{ "type": "error", "requestId": "uuid", "error": "Token expired" }
REST API Endpoints
| Endpoint | Method | Description |
|---|---|---|
/health |
GET | Health check with connection stats |
/console |
GET | Web console UI |
/auth/google |
POST | Initiate OAuth flow |
/auth/google/callback |
GET | OAuth callback |
/api/link/generate |
POST | Generate QR code link token |
/api/devices |
GET | List user's linked devices |
/api/devices/:id |
DELETE | Unlink a device |
/api/mcp/services |
GET | List available MCP services |
Available MCP Tools
Gmail (gmail.*)
| Tool | Description |
|---|---|
gmail.get_unread |
Get unread email count and summaries |
gmail.search |
Search emails by query |
gmail.get_message |
Get full email content by ID |
Calendar (calendar.*)
| Tool | Description |
|---|---|
calendar.get_today |
Get today's events |
calendar.get_events |
Get events for N days |
calendar.create_event |
Create a new event |
Database Schema
-- Users (from Google OAuth)
users (id, google_id, email, display_name, profile_picture_url, created_at, updated_at)
-- Linked devices
devices (id, user_id, device_name, device_model, last_seen_at, last_heartbeat_at, linked_at, created_at)
-- QR code link tokens (single-use, 5 min expiry)
link_tokens (id, user_id, expires_at, used_at, used_by_device_id, created_at)
-- Encrypted OAuth tokens
oauth_tokens (id, user_id, provider, access_token_encrypted, refresh_token_encrypted, ...)
Security Considerations
Token Encryption
OAuth tokens are encrypted at rest using AES-256-GCM. The encryption key comes from the ENCRYPTION_KEY environment variable.
Link Token Security
- Cryptographically random (32 bytes)
- Single-use (marked used after successful link)
- Short expiration (5 minutes)
- Stored as SHA-256 hash (original never stored)
Google API Scopes
This app requests restricted scopes (gmail.readonly, calendar.events). For public deployment, you'll need Google's CASA security assessment ($15K-$75K/year). For testing, keep the app in "Testing" mode (100 user limit).
Future Enhancements
- [ ] Third-party MCP server registration via console
- [ ] Push notifications via FCM
- [ ] Usage analytics and tool popularity metrics
- [ ] Multi-tenancy for organizations
- [ ] PostgreSQL migration for horizontal scaling
Related Documentation
License
MIT
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。