
Clerk MCP Server Template
A production-ready template for building Model Context Protocol servers with Clerk authentication on Cloudflare Workers, allowing AI assistants to securely access user data and business logic in Clerk-authenticated applications.
README
Clerk MCP Server Template
A production-ready template for building Model Context Protocol (MCP) servers with Clerk authentication on Cloudflare Workers. This template provides everything you need to create secure, authenticated MCP tools that integrate with your existing Clerk-powered applications.
Features
- ✅ Clerk Authentication Integration - Complete OAuth 2.0 flow with Clerk
- ✅ Cloudflare Workers - Serverless edge computing with global distribution
- ✅ Durable Objects - Persistent MCP session state management
- ✅ KV Storage - Temporary OAuth session storage
- ✅ Security - HMAC-signed state parameters and automatic token refresh
- ✅ TypeScript - Full type safety throughout the codebase
- ✅ Example Tools - Ready-to-use example MCP tools
- ✅ Development Tools - ESLint, Prettier, and MCP Inspector integration
Why This Template?
This template bridges your existing Clerk-authenticated application with Claude AI through MCP tools. Perfect for:
- SaaS Applications: Give Claude access to your user data and business logic
- Customer Support: Let Claude query your systems with proper user context
- Data Analysis: Provide Claude with authenticated access to your APIs
- Workflow Automation: Create secure, user-specific automations
Quick Start
1. Prerequisites
- Node.js 22.x or later
- A Clerk account with API keys
- A Cloudflare account with Workers enabled
- An existing application using Clerk for authentication
2. Use This Template
git clone https://github.com/your-username/clerk-mcp-template.git my-mcp-server
cd my-mcp-server
npm install
3. Configure Environment Variables
Copy the example environment file:
cp .dev.vars.example .dev.vars
Update .dev.vars
with your Clerk keys and app URL:
CLERK_SECRET_KEY=sk_test_your_actual_clerk_secret_key
CLERK_PUBLISHABLE_KEY=pk_test_your_actual_clerk_publishable_key
APP_URL=https://your-app.com
Important:
APP_URL
should point to your existing Clerk-authenticated application where you'll implement the MCP auth flow.
4. Create KV Namespace
Create a KV namespace for OAuth session storage:
wrangler kv:namespace create "OAUTH_KV"
Update the id
in wrangler.jsonc
with the generated namespace ID.
5. Update Configuration
wrangler.jsonc
:
- Change
name
from"your-mcp-server"
to your desired worker name - Update the KV namespace ID with the one generated above
src/index.ts
:
- Update the server name and version in the
McpServer
constructor - Replace example tools with your own (see examples below)
6. Start Development
npm run dev
The server will be available at http://localhost:8788
Architecture
graph TB
A[MCP Client] --> B[Cloudflare Worker]
B --> C[OAuth Provider]
C --> D[Clerk Authentication]
B --> E[Durable Objects]
B --> F[KV Storage]
B --> G[Your API]
E --> H[MCP Session State]
F --> I[OAuth Sessions]
D --> J[User Authentication]
G --> K[Your Application Data]
Authentication Flow
- MCP Client connects to
/sse
endpoint - OAuth redirect to
/authorize
endpoint - User authentication via Clerk (you implement this part)
- Token exchange at
/callback
endpoint - Session creation in Durable Objects
- MCP tools become available with authenticated context
Integrating with Your Application
Step 1: Add MCP Authentication Route
Create an authentication route in your existing Clerk application at
/auth/mcp
. This route handles the OAuth flow initiated by the MCP server.
React Router v7 (Framework Mode) Example
This example shows integration with React Router v7 in framework mode (formerly Remix), but you can adapt it to Next.js, Express, or any framework.
app/routes/auth.mcp.tsx
:
import { createClerkClient } from '@clerk/express'
import { redirect, type LoaderFunctionArgs } from 'react-router'
export async function loader({ request }: LoaderFunctionArgs) {
const url = new URL(request.url)
const state = url.searchParams.get('state')
const callbackUrl = url.searchParams.get('callback_url')
const clientName = url.searchParams.get('client_name')
if (!state || !callbackUrl) {
throw new Error('Missing required parameters')
}
// Get the authenticated user's session token
const clerkClient = createClerkClient({
secretKey: process.env.CLERK_SECRET_KEY,
publishableKey: process.env.CLERK_PUBLISHABLE_KEY,
})
const clerkAuth = (await clerkClient.authenticateRequest(request)).toAuth()
const sessionToken = await clerkAuth?.getToken()
if (!sessionToken) {
// Redirect to sign-in if not authenticated
const signInUrl = new URL('/sign-in', request.url)
signInUrl.searchParams.set('redirect_url', request.url)
return redirect(signInUrl.toString())
}
// Redirect back to MCP server with token
const redirectUrl = new URL(callbackUrl)
redirectUrl.searchParams.set('clerk_token', sessionToken)
redirectUrl.searchParams.set('state', state)
return redirect(redirectUrl.toString())
}
// Optional: Add a component for showing consent screen
export default function McpAuth() {
return (
<div className="max-w-md mx-auto mt-8 p-6 bg-white rounded-lg shadow-md">
<h1 className="text-xl font-bold mb-4">Authorize MCP Access</h1>
<p className="text-gray-600 mb-4">
Claude AI is requesting access to your account data.
</p>
<p className="text-sm text-gray-500">
This will redirect you automatically...
</p>
</div>
)
}
Step 2: Create API Endpoints
Add protected API endpoints in your application that the MCP server can call with authenticated requests.
app/routes/api.users.tsx
:
import { createClerkClient } from '@clerk/express'
import { json, type LoaderFunctionArgs } from 'react-router'
export async function loader({ request }: LoaderFunctionArgs) {
try {
const clerkClient = createClerkClient({
secretKey: process.env.CLERK_SECRET_KEY,
publishableKey: process.env.CLERK_PUBLISHABLE_KEY,
})
// Verify the request is authenticated
const clerkAuth = await clerkClient.authenticateRequest(request)
const userId = clerkAuth.toAuth()?.userId
if (!userId) {
return json({ error: 'Unauthorized' }, { status: 401 })
}
// Your business logic here
const users = await getUsersForCurrentUser(userId)
return { users }
} catch (error) {
return json({ error: 'Internal server error' }, { status: 500 })
}
}
Step 3: Customize MCP Tools
Replace the example tools in src/index.ts
with your own:
// Custom tool example
this.server.tool(
'getUsers',
'Fetch all users from your application',
{},
this.requireAuth(async () => {
const users = await this.makeApiRequest('api/users')
return {
content: [
{
type: 'text',
text: `Found ${users.length} users:\n${JSON.stringify(users, null, 2)}`,
},
],
}
}),
)
Configuration Reference
Environment Variables
Variable | Description | Required |
---|---|---|
CLERK_SECRET_KEY |
Your Clerk secret key | ✅ |
CLERK_PUBLISHABLE_KEY |
Your Clerk publishable key | ✅ |
APP_URL |
Your application URL | ✅ |
Add your own application-specific environment variables to the Env
interface
in src/types.ts
.
Clerk JWT Templates
Create a JWT template in your Clerk Dashboard for token generation:
- Go to JWT Templates in your Clerk Dashboard
- Create a new template (e.g., "mcp-server")
- Update the template name in
src/index.ts
:
const token = await getToken(
// ... token manager
(this as any).env.CLERK_SECRET_KEY,
'your-template-name', // Update this
)
Development
Available Scripts
npm run dev # Start development server
npm run deploy # Deploy to Cloudflare Workers
npm run inspect # Launch MCP Inspector
npm run lint # Run ESLint + format
npm run typecheck # Run TypeScript type checking
npm run validate # Run typecheck + lint
Testing with MCP Inspector
- Start the development server:
npm run dev
- Open MCP Inspector
- Set transport type to SSE
- Connect to
http://localhost:8788/sse
- Complete the authentication flow
- Test your tools
Deployment
1. Set Production Secrets
wrangler secret put CLERK_SECRET_KEY
wrangler secret put CLERK_PUBLISHABLE_KEY
wrangler secret put APP_URL
2. Create Production KV Namespace
wrangler kv:namespace create "OAUTH_KV" --env production
Update the production KV namespace ID in wrangler.jsonc
.
3. Deploy and Configure
npm run deploy
Add your deployed server to Claude Desktop MCP configuration:
{
"mcpServers": {
"my-app": {
"command": "npx",
"args": [
"@modelcontextprotocol/server-remote",
"https://your-mcp-server.your-subdomain.workers.dev/sse"
]
}
}
}
Project Structure
clerk-mcp-template/
├── src/
│ ├── index.ts # Main MCP server class and tools
│ ├── auth.ts # OAuth authentication handlers
│ ├── clerk.ts # Clerk authentication utilities
│ ├── utils.ts # Utility functions (HMAC, logging, etc.)
│ └── types.ts # TypeScript type definitions
├── wrangler.jsonc # Cloudflare Worker configuration
├── package.json # Dependencies and scripts
├── tsconfig.json # TypeScript configuration
├── eslint.config.js # ESLint configuration
├── .dev.vars # Development environment variables
└── README.md # This file
Security Considerations
- OAuth 2.0 ensures secure authentication flow
- HMAC signatures protect state parameters from tampering
- Automatic token refresh handles session expiration
- Session cleanup removes expired OAuth sessions
- Secure headers include proper CORS and authentication headers
Troubleshooting
Common Issues
Authentication fails:
- Verify Clerk API keys are correct
- Ensure your authentication route is implemented
- Check that JWT template exists in Clerk Dashboard
KV namespace errors:
- Verify namespace ID in
wrangler.jsonc
- Ensure namespace is created and bound
Tools not working:
- Check that user is authenticated
- Verify API endpoints are correct
- Review Cloudflare Workers logs
Debugging
# View real-time logs
wrangler tail
# Check deployment status
wrangler deployments list
# Test locally with debugging
npm run dev
Contributing
- Fork this repository
- Create a feature branch:
git checkout -b feature/amazing-feature
- Commit your changes:
git commit -m 'Add amazing feature'
- Push to the branch:
git push origin feature/amazing-feature
- Open a Pull Request
Resources
- Model Context Protocol Specification
- Clerk Authentication Documentation
- Cloudflare Workers Documentation
- Cloudflare MCP Examples
- React Router v7 Documentation
License
MIT License - see LICENSE file for details.
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。