Cursor

MCP (Model Context Protocol) enables your AI assistant to call external tools and services, expanding its capabilities beyond what it can do on its own. With Cursor MCP Servers on Windows, you can run local MCP servers that provide search, browsing, code execution, and filesystem access to your AI workflows, making complex tasks more efficient and automated.

How to use

You integrate MCP servers by configuring one or more stdio (local) servers that your MCP client can invoke. Each server runs as a separate process and exposes a command interface that the AI can call to perform specific actions such as web browsing, searching, or interacting with GitHub and the filesystem. You can run multiple servers in parallel and select the ones that fit your current task to optimize performance and token usage.

How to install

Prerequisites: to set up MCP servers on Windows, you need Node.js v18 or higher and Cursor IDE 0.46 or newer. You should also have basic command-line knowledge and, for some servers, API keys.

Step 1. Install Node.js (if you don’t have it yet) and verify the versions.

Step 2. Open Cursor and prepare MCP server configurations using the project’s local setup options.

Step 3. Add stdio MCP servers with their commands and arguments as shown in the examples below.

Sequential Thinking MCP Server

{"mcpServers": {
  "sequential_thinking": {
    "command": "npx",
    "args": ["-y", "@modelcontextprotocol/server-sequential-thinking"]
  }
}}

Brave Search MCP Server

{"mcpServers": {
  "brave_search": {
    "command": "npx",
    "args": ["-y", "@modelcontextprotocol/server-brave-search"],
    "env": [{"name": "BRAVE_API_KEY", "value": "YOUR_API_KEY"}]
  }
}}

Puppeteer MCP Server (网页浏览)

{"mcpServers": {
  "puppeteer": {
    "command": "npx",
    "args": ["-y", "@modelcontextprotocol/server-puppeteer"]
  }
}}

GitHub tools MCP Server

{"mcpServers": {
  "github": {
    "command": "npx",
    "args": ["-y", "@modelcontextprotocol/server-github"]
  }
}}

File system tools MCP Server

{"mcpServers": {
  "fs_tools": {
    "command": "npx",
    "args": ["-y", "@modelcontextprotocol/server-fs"]
  }
}}

环境变量配置

Some MCP servers require API keys or other secrets. You can provide these via environment variables in the server configuration. For example, Brave Search requires BRAVE_API_KEY. You can set variables at the server level or as global environment variables for your development environment.

{"env": [{"name": "BRAVE_API_KEY", "value": "YOUR_API_KEY"}]}

故障排除

• MCP服务器状态显示为黄色时，尝试刷新状态、检查命令是否正确、重启Cursor IDE
• 无法找到MCP工具时，确保在Composer或Agent模式下使用并明确指示AI使用MCP工具
• Windows权限问题时，尝试以管理员身份运行终端并检查防火墙设置
• 项目级配置文件不起作用时，尝试使用全局配置文件或通过设置界面添加
• Node.js版本问题时，使用nvm切换到Node.js v18或更高版本

最佳实践

• 按需添加服务器，只保留当前任务需要的MCP服务器以减少资源消耗
• 使用Git Bash代替CMD以获得更好的兼容性
• 安全存储API密钥，不要在共享代码中硬编码密钥
• 定期更新MCP服务器以跟随生态系统的发展
• 为不同项目配置独立的MCP服务器设置

常见问题

以下是常见的问题与简要解答。

参考资源

如需更多示例，请使用你自己的配置信息来扩展服务器集合。

Available tools

sequential_thinking

MCP server that provides sequential thinking capabilities for task planning and stepwise reasoning.

brave_search

MCP server enabling web search via Brave Search API.

puppeteer

MCP server enabling web page rendering and browsing through a headless browser.

github

MCP server to interact with GitHub for repository and issue operations.

fs_tools

MCP server exposing filesystem-related capabilities for reading and writing local files.