MCP tutorials

tutorials
mcp-servers

Using MCP

Connect to Remote MCP Servers

远程 MCP 服务器扩展了 AI 应用的能力，使其不仅局限于本地环境，还可以访问托管在互联网上的工具、服务和数据源。

目前，许多客户端已经支持远程 MCP 服务器，从而实现广泛的集成可能性。本指南以 Claude 为例，展示如何通过远程 MCP 服务器进行连接。

远程 MCP 服务器的关键优势是其可访问性。与需要在每个设备上安装和配置的本地服务器不同，远程服务器只需有网络连接，就能通过任何 MCP 客户端访问。这使它们非常适合：
• 基于 Web 的 AI 应用
• 强调易用性的集成场景
• 需要服务端处理或认证的服务

什么是 Custom Connectors？

Custom Connectors 是 Claude 与远程 MCP 服务器之间的桥梁。它们允许你将 Claude 直接连接到与你工作流程最相关的工具和数据源，使 Claude 能在你常用的软件中运行，并从外部工具的完整上下文中获取洞察。
通过 Custom Connectors，你可以：
• 将 Claude 连接到第三方开发者提供的远程 MCP 服务器
• 自建远程 MCP 服务器，与任何工具集成

Connecting to a Remote MCP Server

略.

Connect to Local MCP Servers

Model Context Protocol (MCP) 服务器通过提供对本地资源和工具的安全、受控访问，扩展了 AI 应用程序的能力。

理解local MCP 服务器

MCP 服务器是运行在你电脑上的程序，通过标准协议为 Claude Desktop 提供特定功能。每个服务器会暴露一组工具，Claude 可以在获得你的批准后使用这些工具。

以 Host Infra Server 为例，它提供以下工具：
• 查看主机的型号
• 查看主机的内存使用情况
• 查看主机的CPU使用情况

实现Host Infra MCP Server

环境依赖：

1. Claude Desktop
2. Node.js
3. uv

代码结构

tools.py

import platform
import os
import psutil
import json

def get_host_info() -> str:
    """get host information

    Returns:
        str : the host information in JSON format
    """

    info: dict[str, str ] = {
        "System": platform.system(),
        "Release": platform.release(),
        "Machine": platform.machine(),
        "Processor": platform.processor()
    }

    info["cpu_count"] = psutil.cpu_count(logical=False)
    info["memory_total"] = __to_gib(psutil.virtual_memory().total)
    return json.dumps(info, indent=4)


def __to_gib(bytes_value):
        return str(round(bytes_value / (1024 ** 3), 2)) + " GiB"

def get_memory_detail_info() -> str:
    """get memory information

    Returns:
        str: The memory information in JSON format
    """
    memory_info = psutil.virtual_memory()

    info = {
        "total": __to_gib(memory_info.total),
        "available": __to_gib(memory_info.available),
        "used": __to_gib(memory_info.used),
        "percent": str(memory_info.percent) + "%"
    }
    return json.dumps(info, indent=4)


def get_cpu_detail_info() -> str:
    """get cpu information

    Returns:
        str: The cpu information in JSON format
    """
    cpu_info = psutil.cpu_times()
    cpu_percent = psutil.cpu_percent(interval=1)
    info = {
        "user": cpu_info.user,
        "system": cpu_info.system,
        "idle": cpu_info.idle,
        "cpu_percent": str(cpu_percent) + "%"
    }
    return json.dumps(info, indent=4)

if __name__ == "__main__":
    print(get_host_info())
    print(get_memory_detail_info())
    print(get_cpu_detail_info())

host_infra_mcp.py

import sys
import os
sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
import tools

from mcp.server.fastmcp import FastMCP
import logging


mcp = FastMCP("host infra mcp server ")
mcp.add_tool(tools.get_host_info, "get_host_info")
mcp.add_tool(tools.get_memory_detail_info, "get_memory_detail_info")
mcp.add_tool(tools.get_cpu_detail_info, "get_cpu_detail_info")

@mcp.tool() # 使用注解注入 tool
def foo() -> str:
    return "Hello from foo!"

def main() -> None:
    mcp.run(transport='stdio')  # or sse
    logging.info("mcp server started")


if __name__ == "__main__":
    main()

config MCP server to Claude Desktop

Access Developer Settings

Configure claude_desktop_config.json

{
  "mcpServers": {
    "host_infra_mcp": {
      "command": "uv",
      "args": [
        "--directory",
        "~/workspaces/mcp_demos/server",
        "run",
        "host_infra_mcp.py"
      ]
    }
  }
}

restart claude_desktop and test

case1

case2

case3

inspector

Build an MCP Client

环境依赖： uv add fastmcp

code structure

client.py

import os
import asyncio
from fastmcp import Client

async def main():
    server_path = os.path.abspath(os.path.join(os.path.dirname(__file__), '../server/host_infra_mcp.py'))
    client = Client(server_path)

    async with client:
        host_info = await client.call_tool("get_host_info")
        print(f"Host Info: {host_info.data}")

        memory_info = await client.call_tool("get_memory_detail_info")
        print(f"Memory Info: {memory_info.data}")

        cpu_info = await client.call_tool("get_cpu_detail_info")
        print(f"CPU Info: {cpu_info.data}")

if __name__ == "__main__":
    asyncio.run(main())

结果

---
Host Info: get_host_infoOutput(result='{\n    "System": "Darwin",\n    "Release": "24.6.0",\n    "Machine": "x86_64",\n    "Processor": "i386",\n    "cpu_count": 12,\n    "memory_total": "18.0 GiB"\n}')

---
Memory Info: get_memory_detail_infoOutput(result='{\n    "total": "18.0 GiB",\n    "available": "1.02 GiB",\n    "used": "1.7 GiB",\n    "percent": "94.3%"\n}')

---
CPU Info: get_cpu_detail_infoOutput(result='{\n    "user": 53391.42,\n    "system": 40173.91,\n    "idle": 501844.89,\n    "cpu_percent": "8.7%"\n}')