MCP Protocol Ecosystem: Server/Client Architecture Deep Dive

MCP Protocol Ecosystem

In the mcp_station project, I analyzed the MCP (Model Context Protocol) ecosystem and built a server from scratch.

MCP Ecosystem Structure

graph TD
    A[MCP Host] --> B[MCP Client]
    B --> C[MCP Server 1 - 파일]
    B --> D[MCP Server 2 - DB]
    B --> E[MCP Server 3 - API]

    C --> F[로컬 파일 시스템]
    D --> G[PostgreSQL]
    E --> H[외부 API]

    A --> I[Claude Desktop]
    A --> J[VS Code Copilot]
    A --> K[커스텀 앱]

MCP Server Implementation

from mcp.server import Server
from mcp.types import Tool, TextContent

server = Server("my-mcp-server")

@server.list_tools()
async def list_tools():
    return [
        Tool(
            name="search_documents",
            description="문서를 검색합니다",
            inputSchema={
                "type": "object",
                "properties": {
                    "query": {"type": "string"}
                },
                "required": ["query"]
            }
        )
    ]

@server.call_tool()
async def call_tool(name: str, arguments: dict):
    if name == "search_documents":
        results = search(arguments["query"])
        return [TextContent(type="text", text=str(results))]

Transport Protocol

sequenceDiagram
    participant H as MCP Host
    participant C as MCP Client
    participant S as MCP Server

    H->>C: 사용자 요청
    C->>S: initialize
    S-->>C: capabilities
    C->>S: tools/list
    S-->>C: 도구 목록
    C->>S: tools/call (search_documents)
    S-->>C: 검색 결과
    C-->>H: 결과 전달

The Value of MCP

MCP has a strong chance of becoming the standard for AI agent ecosystems. Building with it hands-on in mcp_station made that clear.