Simple Streamable HTTP MCP Server
A reference implementation demonstrating proper MCP server patterns with HTTP transport, featuring session management, progress notifications, and example tools for testing server functionality. Serves as a clean template for building MCP servers with streamable responses and comprehensive error handling.
README
Simple Streamable HTTP MCP Server
A clean implementation of an MCP server using the Streamable HTTP transport, following the official MCP patterns.
Features
- Streamable HTTP Transport: Supports both JSON responses and Server-Sent Events (SSE)
- Session Management: Proper session handling with unique session IDs
- Progress Notifications: Long-running tools send progress updates
- Multiple Tools:
hello_world: Simple greeting toolget_server_info: Returns server informationlong_running_test: Configurable duration test with progress (default: 30s)slow_test: 10-minute test with progress updates
Architecture
This server follows the MCP team's recommended patterns:
- Session-based Architecture: Each client gets a unique session ID
- Transport Reuse: Transports are stored and reused for subsequent requests
- Proper Initialization: New sessions are created only on initialization requests
- Clean Shutdown: All transports are properly closed on server shutdown
Running the Server
# Install dependencies
npm ci
# Run the server (default port: 3001, DEBUG logging enabled by default)
npm run start
# Run without debug logging, or edit `.env` to set DEBUG=false
DEBUG=false npm run start
# Run in development mode (auto-reload)
npm run dev
Testing
Use the provided test script to verify the server is working correctly:
npm run test:http
This will test:
- Session initialization
- Tool listing and calling
- Session ID management
- SSE streaming
- Session termination
Progress Notifications
The long_running_test and slow_test tools demonstrate progress notifications:
// Example tool call with progress
{
"jsonrpc": "2.0",
"method": "tools/call",
"params": {
"name": "long_running_test",
"arguments": {
"duration": 60,
"steps": 12,
"message": "Testing progress notifications"
},
"_meta": {
"progressToken": "unique-progress-token"
}
},
"id": 1
}
The server will send progress notifications:
{
"method": "notifications/progress",
"params": {
"progress": 1,
"total": 12,
"progressToken": "unique-progress-token"
}
}
Key Differences from Basic Examples
- Proper Session Management: Sessions are created per client, not per request
- Transport Lifecycle: Transports persist across requests within a session
- Error Handling: Comprehensive error handling for all edge cases
- Progress Support: Long-running operations can report progress
- Clean Architecture: Clear separation between server creation and request handling
Environment Variables
PORT: Server port (default: 3001)DEBUG: Enable debug logging (trueby default)
Endpoints
POST /mcp: Handle JSON-RPC requestsGET /mcp: Handle SSE streamingDELETE /mcp: Terminate sessionsGET /health: Health check endpoint
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。