A Deep Dive: How to Use MCP and Skills in Trae, Cursor and Claude Code

目录

As AI coding tools evolve, plain “conversation” no longer meets complex development needs. We need AI to be able to proactively “do things”—like query a database, read local files, or directly run scripts. That’s where MCP (Model Context Protocol) and Skills come in.

This article walks you through configuring and using these powerful capabilities in the three hottest AI coding tools right now—Trae, Cursor and Claude Code.

What is MCP (Model Context Protocol)?

Simply put, MCP is an open standard introduced by Anthropic; it’s like the AI’s “USB port”.

• MCP Server: provides tools or data (for example a service that checks the weather, or one that can read and write GitHub).
• MCP Client: the AI software that uses those tools (for example Trae, Claude Desktop, Cursor).

With MCP, you can turn any external tool into the AI’s “hands and feet.”

---

1. Trae IDE: Native and Powerful Skills Support

Trae is an AI IDE from ByteDance; its MCP support is very intuitive and it introduces a unique Skills concept.

How to Configure an MCP Server

1. Open settings: click the gear icon at the bottom left, or use Cmd/Ctrl + ,.
2. Go to MCP management: find “Features” -> “MCP Servers” in the settings menu.
3. Add a service:
   • Trae usually offers some presets (such as Local Shell, Git, etc.).
   • Click “Add Custom Server” and enter the launch command.
   • Example (adding SQLite support):
     • Name: sqlite
     • Command: uvx (make sure uv is installed) or npx
     • Args: mcp-server-sqlite --db-path ./my-database.db

How to Use Skills

Trae’s Skills are a higher-level agent capability. Besides the basic MCP tools, Trae also lets you load dedicated Skill packages (for example a UI design expert or a code review expert).

• How to invoke: in the Chat interface, use natural-language instructions directly, or use the / command to view available skills.
• Skill Creator: Trae even lets you define your own skills via the skill-creator tool, packaging frequently used prompts and workflows into a Skill.

2. Claude Code: The Geek’s Terminal Weapon

Claude Code is Anthropic’s command-line AI assistant; it is the standard reference implementation of MCP and is very flexible to configure.

Installation and Initialization

Make sure you’ve installed the tool via npm install -g @anthropic-ai/claude-code.

Adding MCP via the CLI

Claude Code provides convenient command-line interaction for managing MCP.

# Start the configuration wizard
claude mcp add

# Or add it with a single command (for example adding GitHub support)
claude mcp add github --scope user -- npx -y @modelcontextprotocol/server-github

• --scope user: makes the tool visible to all projects.
• --scope project: available only to the current project (generates a .mcp.json file).

Editing the Config File Manually

If you like to be in control of everything, you can edit the config file directly:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Example config:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/me/Desktop"]
    }
  }
}

Once configured, type /mcp in Claude Code to see the connection status.

---

3. Cursor: The Art of Context and Rules

Cursor’s current core logic leans more on Context (@Symbols) and Rules, but it is gradually catching up on MCP support.

The Core Alternatives: Docs and CursorRules

Although Cursor’s MCP support is still evolving, its existing mechanisms are just as powerful:

1. @Docs (documentation skill):
   • In Chat, type @Docs -> Add new doc.
   • Paste the library’s documentation link and Cursor indexes it automatically. This is like temporarily installing a “knowledge-base Skill” for the AI.
2. CursorRules (behavior skill):
   • Create a .cursorrules file in the project root.
   • Define rules in natural language, for example: “When writing React components, always use TypeScript and Tailwind CSS.” This is like hardening a “code-standards expert Skill.”

Enabling MCP (Beta)

If your Cursor version supports MCP:

1. Go to Settings -> Features (or Beta).
2. Look for the MCP tab.
3. Adding is similar to Trae: specify the Server’s Command and Args.

---

Hands-On: Add a “File Search” Skill to Your AI

Whichever tool you use, the most practical starter MCPs are fetch and filesystem. Here we use Trae as an example to add a skill that can fetch web-page content:

1. Prepare the environment: make sure Node.js is installed.
2. Configure the Server:
   • Command: npx
   • Args: -y @modelcontextprotocol/server-fetch
3. Use it:
   • In Trae Chat, type: “Please fetch the content of https://example.com and summarize it.”
   • Trae recognizes the intent automatically, calls the fetch tool to get the content, and then generates the summary for you.

Summary

• Trae: offers the friendliest GUI and a native Skills extension system, ideal for developers who want things to work out of the box.
• Claude Code: ideal for CLI lovers and advanced users who need to strictly follow the MCP standard.
• Cursor: currently better at leveraging local context and documentation indexing, ideal for heavy code completion and refactoring.

Once you master MCP and Skills, your AI is no longer just a chatbot but a super-assistant that can really “roll up its sleeves and work.”

中英对照

文章作者：阿文

文章链接： https://www.awen.me/post/bee0dc51.html

版权声明：本博客所有文章除特别声明外，均采用 CC BY-NC-SA 4.0 许可协议。转载请注明来自 阿文的博客！