mcp-server-blog
A demo MCP server illustrating production patterns including OAuth 2.0 token validation, workspace context management, Firestore integration, and Zod schema validation using TypeScript.
README
MCP Server Demo: OAuth, TypeScript & Firestore Patterns
This repository provides a stripped-down, illustrative implementation of a Model Context Protocol (MCP) server. It showcases the architectural patterns, security considerations (OAuth 2.0), and development practices (TypeScript, Firestore, Zod) detailed in our accompanying article:
➡️ Read the full story: "Building a Production-Ready MCP Server with OAuth, TypeScript, and Our Battle Scars"
This demo focuses on the MCP Resource Server component and assumes you have a separate OAuth 2.0 Authorization Server.
Update: MCP SDK version 1.12.0 introduced the Authorization Server Metadata (/.well-known/oauth-authorization-server) support and removes the need to proxy the oauth calls through the MCP resource server.
Purpose
This repository is intended as a learning resource to:
- Demonstrate a practical implementation of an MCP server using the TypeScript MCP SDK.
- Illustrate how to integrate OAuth 2.0 for securing tool calls.
- Showcase patterns for managing workspace context and multi-tenancy.
- Provide examples of using Zod for schema definition and validation.
- Offer insights into structuring tools, using Higher-Order Components (HOCs) for common logic, and interacting with Firestore.
This is NOT a production-ready, plug-and-play server for all use cases. It omits specific business logic and assumes a pre-existing OAuth Authorization Server.
Key Features & Patterns Demonstrated
- MCP TypeScript SDK Integration: Core server setup and tool registration.
- OAuth 2.0 Token Validation: Securely handling Bearer tokens (via the SDK middleware).
- Workspace Context Management:
- Explicit
workspace_idin tool arguments. withWorkspaceAccessHigher-Order Component (HOC) for authentication and workspace authorization.
- Explicit
- Firestore Integration:
- Fetching user data, OAuth token information (conceptual), and tool-specific data.
- Using Firestore emulators for local development and testing.
- Zod for Schemas & Validation: Defining input schemas for tools and leveraging Zod for runtime validation.
- Type-Safe Development: Leveraging TypeScript for robust code.
- Utility Functions: Examples of
fetchResourceListfor DRY data fetching. - Standardized Error Handling: Using
throw new Error()for clear error propagation. - Example Tool Structures: Basic tool definitions showcasing the patterns.
- Environment Variable Configuration: For database and OAuth settings.
Architectural Overview
This demo represents the MCP Resource Server. It expects OAuth 2.0 Bearer tokens issued by a separate OAuth Authorization Server.
[Client / LLM with MCP Client SDK]
|
| (HTTPS Request with Bearer Token)
v
[This MCP Resource Server (Node.js / TypeScript)]
| 1. MCP SDK Middleware (parses request, extracts token)
| 2. `withWorkspaceAccess` HOC
| a. Using the userId associated with the token
| b. Validates the user can access the workspace_id in the request
| 3. Tool Handler Execution (interacts with Firestore based on validated context)
|
v
[Google Firestore (Database)]
The OAuth Authorization Server (which you would provide or have existing) is responsible for:
- Authenticating users.
- Issuing OAuth tokens (Access Tokens, Refresh Tokens).
- Managing OAuth clients (Dynamic Discovery).
This Resource Server then validates the tokens received from clients.
Prerequisites
- Node.js (v18.x or later recommended)
- npm or yarn
- Access to a Google Cloud Project with Firestore enabled OR Google Cloud SDK configured for Firestore Emulator.
- An existing OAuth 2.0 Authorization Server.
Getting Started
-
Clone the repository:
git clone https://github.com/portal-labs-infrastructure/mcp-server-blog cd mcp-server-blog -
Install dependencies:
npm install # or yarn install -
Set up Environment Variables: Copy the
.env.examplefile to a new file named.env:cp .env.example .envNow, edit
.envand fill in the required configuration values:# Firestore Configuration # If using Firestore Emulator, these might not all be strictly needed, # but ensure your gcloud CLI is configured or provide necessary emulator host. PROJECT_ID="your-gcp-project-id" # FIRESTORE_EMULATOR_HOST="localhost:8081" # Uncomment if using emulator and not relying on gcloud config # OAuth 2.0 Configuration (for this Resource Server to validate tokens) # This depends on your OAuth Authorization Server's setup. OAUTH_ISSUER_URL="https_your_auth_server_com" # MCP Server Configuration BASE_URL="http://localhost:8080" # URL this server is accessible atImportant: The OAuth configuration is crucial. This server needs to know how to validate tokens issued by your Authorization Server. Consult your Auth Server's documentation.
-
(Optional) Seed Firestore Data: If you have seed scripts or want to manually add some sample users, OAuth tokens (matching what your Auth server would issue), and workspace data to your Firestore instance/emulator, do so now. This will make testing the tools more meaningful.
Running the Server
- Development Mode (with Nodemon for auto-restarts):
npm run dev - Production Mode:
bash npm run build npm startThe server will typically start onhttp://localhost:8080(or the port specified in.env).
Running with Firestore Emulator
- Ensure Google Cloud SDK is installed and configured.
- Start the Firestore emulator in a separate terminal:
(Adjust port if needed and updategcloud emulators firestore start --host-port=localhost:8081FIRESTORE_EMULATOR_HOSTin.envor ensure your application automatically detects it viagcloudenvironment variables). - Run the MCP server as described above. It should connect to the emulator.
Key Patterns & Concepts Demonstrated in Code
Look for these patterns in the src directory:
src/index.ts: Main MCP server setup.src/controllers/mcpController.ts: Tool registration and MCP controller handling incoming requests.src/services/: Service layer for handling Firestore interactions.src/tools/: Example tool definitions.- Each tool will have an
inputSchema(Zod) and ahandler. - Handlers will likely be wrapped with
withWorkspaceAccess.
- Each tool will have an
src/utils/withWorkspaceAccess.ts: The Higher-Order Component for workspace checks.src/utils/fetchResourceList.ts: Example of reusable data fetching utility.src/utils/types.ts: Shared TypeScript types and Zod schemas (e.g., forEntityType,ResourceType).
Directory Structure
.
├── src/
│ ├── tools/ # Tool definitions
│ │ ├── getAgentTool.ts
│ │ └── ...
│ ├── utils/ # Shared utilities, HOCs, types
│ │ ├── withWorkspaceAccess.ts
│ │ ├── types.ts
│ │ └── ...
│ ├── services/ # Interaction logic
│ │ ├── firestoreService.ts # Firestore interaction logic
│ │ └── ...
│ ├── config/ # Configuration loading
│ └── index.ts # Main server setup
├── .env.example # Example environment variables
├── .env # Your local environment variables (ignored by git)
├── package.json
├── tsconfig.json
└── ...
What This Demo Is (and Isn't)
- IS: A demonstration of server-side patterns for building a secure, multi-tenant MCP Resource Server.
- IS: A way to see TypeScript, Zod, Firestore, and OAuth concepts applied in an MCP context.
- IS NOT: A complete, production-ready OAuth Authorization Server (you need to provide that).
- IS NOT: A library or SDK to be directly consumed (it's an example application).
- IS NOT: Filled with complex business logic (tools are illustrative).
Contributing
This is primarily a demo repository. However, if you find bugs or have suggestions for improving the clarity of the demonstrated patterns, feel free to open an issue or submit a pull request.
License
This project is licensed under the MIT License - see the LICENSE file for details.
Acknowledgements
This demo is heavily inspired by the experiences and patterns discussed in the article: "Building a Production-Ready MCP Server with OAuth, TypeScript, and Our Battle Scars".
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。