MCP 도구 및 MCP 서버로 사용자 관리 에이전트 구축을위한 Python 구현

작업 개요

간단한 도구로 MCP 서버를 만들고 실행합니다. MCP 클라이언트와 간단한 사용자 관리 에이전트를 구현하여 만든 서버에서 MCP 도구를 사용할 수 있습니다.

학습 목표

이 프로젝트와 함께 탐구하고 일함으로써, 당신은 배울 것이다:

• 간단한 MCP 서버를 구성하는 방법
• 클라이언트를 구성하고 MCP 서버에 연결하는 방법
• MCP 서버에서 도구로 간단한 에이전트를 만드는 방법
• MCP의 주요 특징

️ 건축

₢ 킹

요구사항

-Python: 3.11 이상

• ** 통화**: requirements.txt에 나열
• 도커 및 도커 컴스
• 포스트맨

❤️ 작업:

주요 지점에서 작업이 당신을 위해 열심히하면, 다음으로 전환 with-detailed-description 지점

사용자 관리 에이전트를 구현해야 합니다. User Management Service 내에서 CRUD 작업을 수행할 수 있습니다.

만들기 및 실행 MCP 서버:

1. 루트 docker-compose 실행 (이전 작업에서 가지고있는 경우 선택 단계)
2. mcp server 열기
3. 모든 *** TODO *** 구현
4. mcp server 실행

OPTIONAL: Postman에서 MCP 서버 작업

1. Postman에 mcp.postman collection 가져 오기
2. 명세.

Framework-based MCP (Server & Client)

Python implementation for building Users Management Agent with MCP tools and MCP server

🎯 Task Overview

Create and run MCP server with simple tools. Implement simple Users Management Agent with MCP Client that will use MCP tools from created server.

🎓 Learning Goals

By exploring and working with this project, you will learn:

• How to configure simple MCP server
• How to configure client and connect to MCP server
• How to create simple Agent with tools from MCP server
• Key features of MCP

🏗️ Architecture

task/
├── agent/
│   ├── models/           
│   │   └──message.py         ✅ Complete
│   ├── app.py                🚧 TODO: implement logic
│   ├── openai_client.py      🚧 TODO: implement logic
│   └── mcp_cleint.py         🚧 TODO: implement logic
└── mcp_server/               
    ├── server.py             🚧 TODO: implement logic
    ├── user_client.py        ✅ Complete
    └── Dockerfile            ✅ Complete

📋 Requirements

• Python: 3.11 or higher
• Dependencies: Listed in requirements.txt
• Docker and Docker Compose
• Postman

✍️ Tasks:

If the task in the main branch is hard for you, then switch to the with-detailed-description branch

You need to implement the Users Management Agent, that will be able to perform CRUD operations within User Management Service.

Create and run MCP server:

1. Run root docker-compose (Optional step in case if you have it from previous tasks)
2. Open mcp_server
3. Implement all TODO
4. Run mcp_server

OPTIONAL: Work with MCP server in Postman

1. Import mcp.postman_collection to Postman
2. Make init call and get mcp-session-id in response headers
3. Make init-notification. Pay attention that you need to use mcp-session-id retrieved from init request. it should return 202 status
4. Get tools (don’t forget about mcp-session-id). It should return stream with tools.
5. Call calculator (don’t forget about mcp-session-id). It should return stream tool execution result.

Create and run Agent:

1. Open mcp_client and implement all TODO
2. Open openai_client and implement all TODO
3. Open prompts and write System prompt
4. Open app and implement all TODO
5. Run application mcp_client and test that it is connecting to MCP Server and works properly
6. Try with your solution with fetch MCP https://remote.mcpservers.org/fetch/mcp and check the differences on the init step (what they have and don’t)

OPTIONAL: Support both (users-management and fetch) MCP servers:

1. Remember that we have 1-to-1 connection between MCP client and MCP server!
2. You need to think of the way how to change current flow to support tools from different MCP servers and implement it