MCP Server Template
A production-ready TypeScript template for building MCP servers with dual transport support (stdio/HTTP), OAuth 2.1 foundations, SQLite caching, observability, and security features including PII sanitization and rate limiting.
README
MCP Server Template
A production-ready template for building Model Context Protocol (MCP) servers with TypeScript.
Features
- MCP SDK 1.24.3 - Latest SDK with 2025-11-25 spec support
- Dual Transport - Stdio (Claude Desktop) and HTTP (cloud deployment)
- OAuth 2.1 Foundations - Protected resource metadata, bearer token structure
- SQLite Caching - TTL-based caching with sql.js (WebAssembly)
- Observability - Sentry error tracking + OpenTelemetry tracing
- Security - PII sanitization, rate limiting, DNS rebinding protection
- Type Safety - Strict TypeScript with Zod validation
Quick Start
# Clone and install
git clone <this-repo> my-mcp-server
cd my-mcp-server
npm install
# Development mode (hot reload)
npm run dev
# Build and run
npm run build
npm start
# Test with MCP Inspector
npm run inspector
Project Structure
src/
├── index.ts # CLI entry point
├── server.ts # MCP server implementation
├── instrumentation.ts # Sentry + OpenTelemetry setup
├── config/
│ └── index.ts # Environment configuration
├── db/
│ ├── database.ts # SQLite connection (sql.js)
│ └── cache.ts # TTL-based caching
├── tools/
│ ├── registry.ts # Tool registration pattern
│ └── examples.ts # Example tool implementations
├── transport/
│ └── http-transport.ts # HTTP with OAuth foundations
├── shared/
│ ├── logger.ts # Structured logging
│ ├── errors.ts # Custom error classes
│ ├── rate-limiter.ts # Per-source rate limiting
│ ├── pii-sanitizer.ts # PII detection/removal
│ └── tracing.ts # OpenTelemetry utilities
└── types/
└── index.ts # TypeScript type definitions
Configuration
Environment variables (prefix with MCP_SERVER_):
| Variable | Default | Description |
|---|---|---|
MCP_SERVER_NAME |
mcp-server-template |
Server name |
MCP_SERVER_VERSION |
0.1.0 |
Server version |
MCP_SERVER_LOG_LEVEL |
info |
Log level: debug, info, warning, error |
MCP_SERVER_DB_PATH |
.data/cache.db |
SQLite database path |
MCP_SERVER_CACHE_ENABLED |
true |
Enable/disable caching |
MCP_SERVER_CACHE_TTL |
3600 |
Default cache TTL (seconds) |
MCP_SERVER_TIMEOUT |
30000 |
Request timeout (ms) |
MCP_SERVER_TRANSPORT |
stdio |
Transport: stdio or http |
MCP_SERVER_PORT |
3000 |
HTTP port (when transport=http) |
MCP_SERVER_HOST |
127.0.0.1 |
HTTP host (when transport=http) |
MCP_SERVER_SENTRY_DSN |
- | Sentry DSN for error tracking |
OTEL_ENABLED |
false |
Enable OpenTelemetry tracing |
OTEL_EXPORTER_OTLP_ENDPOINT |
- | OTLP collector endpoint |
MCP_SERVER_DEBUG |
false |
Debug mode (skips auth) |
Creating Tools
Tools are registered with the ToolRegistry using Zod schemas:
import { z } from 'zod';
import { getToolRegistry } from './tools/registry.js';
// Define input schema
const MyToolInputSchema = z.object({
query: z.string().min(1).describe('Search query'),
limit: z.number().positive().optional().default(10),
});
type MyToolInput = z.infer<typeof MyToolInputSchema>;
// Implement handler
async function myToolHandler(input: MyToolInput) {
// Your implementation here
return {
success: true,
data: { results: [] },
};
}
// Register tool
const registry = getToolRegistry();
registry.register(
'my_tool',
'Description of what this tool does',
MyToolInputSchema,
myToolHandler
);
Transport Modes
Stdio (Default)
For Claude Desktop and local integrations:
npm start
Add to Claude Desktop config:
{
"mcpServers": {
"my-server": {
"command": "node",
"args": ["/path/to/dist/index.js"]
}
}
}
HTTP
For cloud deployment:
MCP_SERVER_TRANSPORT=http npm start
Endpoints:
GET /health- Health checkGET /.well-known/mcp- MCP server metadataGET /.well-known/oauth-protected-resource- OAuth metadata (RFC 9728)GET /mcp- SSE stream for server eventsPOST /mcp- JSON-RPC requestsDELETE /mcp- Close session
Security Features
PII Sanitization
Automatically detects and masks sensitive data:
import { sanitizePii } from './shared/pii-sanitizer.js';
const safe = sanitizePii('Contact: user@example.com');
// Output: "Contact: [EMAIL]"
Rate Limiting
Per-source rate limiting with exponential backoff:
import { getRateLimiter } from './shared/rate-limiter.js';
const limiter = getRateLimiter();
limiter.configure('external-api', {
requestsPerWindow: 100,
windowMs: 60000,
});
await limiter.waitForSlot('external-api');
// Make request...
DNS Rebinding Protection
HTTP transport validates Host headers against allowlist.
Observability
Sentry
Error tracking with PII filtering:
MCP_SERVER_SENTRY_DSN=https://xxx@sentry.io/xxx npm start
OpenTelemetry
Distributed tracing:
OTEL_ENABLED=true \
OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4318 \
npm start
Use tracing utilities:
import { withSpan, createApiSpan } from './shared/tracing.js';
const result = await withSpan('my-operation', async (span) => {
span.setAttribute('custom.attr', 'value');
return doWork();
});
Development
# Type check
npm run check
# Lint
npm run lint
npm run lint:fix
# Format
npm run format
# Test
npm test
npm run test:coverage
# Validate (all checks)
npm run validate
MCP 2025-11-25 Spec Compliance
| Feature | Status |
|---|---|
| Tools | ✅ Implemented |
| Resources | 📝 Scaffolded |
| Prompts | 📝 Scaffolded |
| Streamable HTTP | ✅ Implemented |
| .well-known/mcp | ✅ Implemented |
| OAuth 2.1 Foundations | ✅ Scaffolded |
| Tasks | ❌ Not yet |
| Elicitation | ❌ Not yet |
OAuth 2.1 Implementation
The template includes foundations for OAuth 2.1 per the MCP spec:
- Protected Resource Metadata (RFC 9728) at
/.well-known/oauth-protected-resource - Bearer token middleware structure (implement JWT validation)
- WWW-Authenticate headers with resource_metadata reference
- Scope checking structure for tool authorization
To complete OAuth integration:
- Choose an authorization server (Auth0, Logto, etc.)
- Implement JWT validation in
bearerAuthMiddleware - Add JWKS fetching and caching
- Configure scopes per tool
See MCP Authorization Spec for details.
License
MIT
References
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。