MCP SERVER KB

🚀 OpenAI and MCP Integration

This section demonstrates how to integrate the Model Context Protocol (MCP) with OpenAI's API to create a system that enables OpenAI to access and utilize the tools provided by your MCP server.

📋 Overview

This example illustrates how to:

1. Create an MCP server that exposes a knowledge base tool.
2. Connect OpenAI to this MCP server.
3. Allow OpenAI to dynamically use tools when responding to user queries.

📞 Communication Method

This example uses Server-Sent Events (SSE) transmission for communication between the client and the server, which means:

• The client and the server run in different processes.

🔄 Data Flow Explanation

1. User Query: A user sends a query to the system (e.g., "What is our company's vacation policy?").
2. OpenAI API: OpenAI receives the query and the available tools provided by the MCP server.
3. Tool Selection: OpenAI decides which tools to use based on the query.
4. MCP Client: The client receives the tool invocation request from OpenAI and forwards it to the MCP server.
5. MCP Server: The server executes the requested tool (e.g., retrieves knowledge base data).
6. Response Flow: The tool results are returned to OpenAI via the MCP client.
7. Final Response: OpenAI generates and outputs an answer to the user's query based on the results.

🛠️ How OpenAI Executes Tools

• Tool Registration: OpenAI registers tools in the function call format so that they can be invoked by the API.
• Tool Selection: OpenAI selects the most appropriate tool based on the query content and context.
• Tool Execution: OpenAI invokes the selected tool and processes the returned results.

🌟 The Role of MCP

• Standardization: MCP provides a unified interface for seamless integration of data from different sources.
• Abstraction: MCP hides the complexity of data sources and offers a simple and easy-to-use tool interface.
• Security: MCP ensures data access security through authentication and authorization mechanisms.
• Flexibility: MCP supports multiple data formats and protocols to adapt to different application scenarios.

🛠️ Implementation Details

🖥️ Server

# Build and run the MCP server image using Docker
docker build -t mcp-server .
docker run -p 8000:8000 mcp-server

📱 Client

# Example code: Use OpenAI's Python library to call the MCP server
from openai import Client

client = Client("your_api_key")

response = client.chat.completions.create(
    model="gpt-4",
    messages=[
        {"role": "user", "content": "What is our vacation policy?"}
    ],
    tools=[{
        "type": "function",
        "function": {
            "name": "retrieve_policy_document",
            "parameters": {"policy_type": "vacation"},
            "description": "Retrieve the specified policy document from the knowledge base."
        }
    }]
)

📄 Knowledge Base File

{
  "policies": [
    {
      "id": "1",
      "type": "vacation",
      "content": "Full-time employees are entitled to 20 days of paid vacation per year.\nVacation can be used after 6 months of service.\nUnused vacation days can be carried over to the next year, but no more than 5 days.\nLeave requests must be submitted through the HR system two weeks in advance."
    },
    {
      "id": "2",
      "type": "remote_work",
      "content": "The company adopts a hybrid work model.\nEmployees can work remotely for a maximum of 3 days per week.\nRemote work requires team coordination and approval from the direct supervisor.\nAll remote work requires a stable network connection and a dedicated workspace."
    }
  ]
}

🚀 Example Run Steps

1. Build the MCP Server Image

   docker build -t mcp-server .
   

2. Start the MCP Server Container

   docker run -p 8000:8000 mcp-server
   

3. Install Dependencies and Set Environment Variables

   pip install openai
   export OPENAI_API_KEY="your_api_key"
   

4. Run the Example Code

   python client.py
   

📤 Output Result

{
  "id": "chatcmpl-abcdefgh",
  "message": {
    "role": "assistant",
    "content": "Our company's vacation policy is as follows:\n\n- Full-time employees are entitled to 20 days of paid vacation per year.\n- Vacation can be used after 6 months of service.\n- Unused vacation days can be carried over to the next year, but no more than 5 days.\n- Leave requests must be submitted through the HR system two weeks in advance."
  }
}

This integration example demonstrates how to leverage the powerful capabilities of MCP and OpenAI to build an intelligent question-answering system, allowing users to easily obtain the information they need.