MCP Servers

MCP (Model Context Protocol) is a protocol that allows AI to connect to external tools and data sources. Through MCP, you can give Carry Code access to more features.

---

What is MCP?

MCP (Model Context Protocol) is a standardized AI tool extension protocol. It allows:

• 🔍 Search - Access search engines for real-time information
• 🗄️ Databases - Connect to databases to query data
• 📁 File System - More flexible file access
• 🌐 API Calls - Call various external services
• 🛠️ Custom Tools - Add your own tools

---

Adding an MCP Server

Step 1: Open MCP Management

Type /mcp and press Space

Step 2: Select Add

Select add

Step 3: Choose Connection Method

MCP servers support three connection methods:

Method 1: SSE (Server-Sent Events)

For remote services:

Server Name: my-mcp-server
Server URL: https://mcp.example.com/sse
Description: This is a sample MCP server

Method 2: Stdio

For local programs:

Server Name: local-tool
Command: node
Arguments: /path/to/mcp-server/index.js

Method 3: StreamableHttp

For HTTP streaming:

Server Name: http-mcp
URL: https://mcp.example.com/mcp

---

Managing MCP Servers

View All Servers

/mcp

Displays all configured MCP servers and their status.

Edit Server

/mcp edit <server-name>

Delete Server

/mcp delete <server-name>

Test Connection

/mcp test <server-name>

---

Recommended MCP Servers

File System

Server	Function	Install Command
@modelcontextprotocol/server-filesystem	File system access	npm install -g @modelcontextprotocol/server-filesystem

Example:

Server Name: filesystem
Command: npx
Arguments: @modelcontextprotocol/server-filesystem /your/project/path

Brave Search

Server	Function	Requirement
@modelcontextprotocol/server-brave-search	Real-time web search	Brave API Key required

Databases

Server	Function	Requirement
@modelcontextprotocol/server-postgres	PostgreSQL	Database connection info required
@modelcontextprotocol/server-sqlite	SQLite	Local SQLite file path

GitHub

Server	Function	Requirement
@modelcontextprotocol/server-github	GitHub API	GitHub Token required

---

MCP Configuration Examples

Local File System

{
  "mcpServers": {
    "filesystem": {
      "type": "stdio",
      "command": "npx",
      "args": ["@modelcontextprotocol/server-filesystem", "/path/to/your/project"]
    }
  }
}

Remote SSE Service

{
  "mcpServers": {
    "remote-search": {
      "type": "sse",
      "url": "https://mcp.example.com/sse"
    }
  }
}

---

Using MCP Tools

After adding an MCP server, AI can automatically use these tools:

1. Search tools - Let AI search for real-time information
2. Database tools - Query data
3. File tools - Access specified directories

---

FAQ

Q: Does the MCP server need to run continuously?

Yes, it needs to keep running to be used. Remote servers need to stay online, local programs need to keep the process running.

Q: Will AI automatically use MCP tools after adding?

Yes. AI will automatically discover and use available MCP tools.

Q: Is MCP secure?

MCP tools require your authorization to use. Recommendations:

• Only add trusted MCP servers
• Protect your API keys
• Regularly check added servers

Q: What’s the difference between MCP and built-in tools?

Feature	Built-in Tools	MCP Tools
Quantity	Limited	Extensible
Function	Basic	Custom
Addition	Ready to use	Requires configuration

---

Next Steps

• Skills - Use preset skills
• Rules - Set coding standards
• AGENTS.md - Project-specific rules