Openai Sdk Knowledge Org

🚀 OpenAI SDK Knowledge MCP Server (unofficial)

An MCP server that thoroughly understands the OpenAI API, built with 100% TypeScript and powered by RAG to answer your technical questions.

🚀 Quick Start

This is an MCP server with in - depth knowledge of the OpenAI API. It is fully built with TypeScript using OpenAI Agents SDK, Hono, Cloudflare Workers, and Drizzle ORM. Powered by RAG, it's ready to address your technical queries.

✨ Features

• Cloudflare stack: Fully utilizes Cloudflare Workers, Queues, D1, Vectorize, and AI Gateway.
• Streamable HTTP MPC Server: Compatible with all MPC clients.
• ChatGPT Deep Research connector: Meets the requirements of ChatGPT's Deep Research connector.
• Always updated: Continuously fetches new content from OpenAI repos and community forums.
• Rapidly built with AI: Developed in tandem with various AI coding tools.

💻 Usage Examples

Basic Usage

Streamable HTTP MCP Server

Use a publicly accessible URL (e.g., ngrok, Cloudflare Tunnel) to serve the endpoints for MCP clients. You can generate the token on the top page:

{
  "mcpServers": {
    "openai-sdk-knowledge.org": {
      "type": "streamable-http",
      "url": "https://openai-sdk-knowledge.org/mcp",
      "headers": {
        "Authorization": "Bearer {your api key here}"
      }
    }
  }
}

For example, you can add this MCP server to Cursor. This MCP server is not limited to Cursor and can be used with any other tools supporting MCP server connections.

OpenAI Responses API's Hosted MCP Server Tool

You can pass https://openai-sdk-knowledge.org/mcp along with a valid API token. Then, you can call the tool in the conversation with the Responses API agent.

ChatGPT Deep Research MCP Connector

For ChatGPT Deep Research customer connector, use the same URL. When the ChatGPT server accesses this app's MCP server endpoint, it returns search and fetch tools as well (see the documentation for details).

📦 Installation

Run Locally

# Clone and install
git clone https://github.com/seratch/openai-sdk-knowledge-org.git
cd openai-sdk-knowledge-org/
npm install

# Configure (add your OpenAI API key)
cp .dev.vars.example .dev.vars
# Edit .dev.vars: OPENAI_API_KEY=sk-your-key-here

# Run it
npm run dev

You can access http://localhost:8787 to see how it works.

Requirements: Node.js 22+ and API keys (OpenAI, GitHub)

🔧 Technical Details

This app is essentially a simple web app running on Cloudflare Workers. The web app provides MCP server protocol - compatible endpoints and a web user interface. For the RAG data pipeline, it collects data from sources and generates asynchronous tasks to run and enqueue them into Cloudflare’s Queue.

src/
├── agents/          # Internally used agents built with OpenAI Agents SDK
├── pipeline/        # RAG data collection and processing
├── server/mcp/      # MCP protocol implementation
├── server/web/      # Web app implementation
├── storage/         # Vector database (Vectorize) and D1 database access
└── index.ts         # App entry point

📄 License

This project is licensed under the MIT License.