🔌 MCP Usage Guide
Empower your AI skills to not just "think" but also "do"
What is MCP?
MCP (Model Context Protocol) is an open protocol that enables AI skills to "take action". Just like mobile apps need permissions to access your photos or location, AI skills need MCP to read files, search the web, and query databases.
❌ Skills Without MCP
You: "Analyze sales.csv for me"
AI: "Please paste the data to me..."
Can only generate text, cannot manipulate files
✅ Skills With MCP
You: "Analyze sales.csv for me"
AI reads file → analyzes data → outputs report
Can read files, search web, query databases
When Do I Need MCP?
When you see a badge like "🔧 requires read_file" on a skill detail page, it means the skill needs MCP to function fully.
For example, "Data Analyst" needs to read your Excel files, "Code Review" needs to read your code files.
AI knowledge has a cutoff date. When fresh information is needed, skills use MCP to search the web.
Platform Configuration Guides
Select your AI tool to view MCP configuration steps
🤖 Configure MCP in Claude Code
~/.claude/claude.json (macOS / Linux) %USERPROFILE%\.claude\claude.json (Windows)
Add to the config file (or merge with existing mcpServers）：
{
"mcpServers": {
"filesystem": {
"command": "npx",
"args": ["-y", "@anthropic/mcp-server-filesystem", "/your/workspace/path"]
},
"fetch": {
"command": "uvx",
"args": ["mcp-server-fetch"]
}
}
}Restart Claude Code, then type "list files in my workspace" — if a file list is returned, MCP is working ✅
MCP Security & Permissions
Different MCP tools have different permission levels. Understanding these helps you evaluate whether a skill is safe and reliable.
read_file — can only read files you specify
web_search — can only search public web pages
list_directory — can only list directory contents
fetch_url — access URLs you specify
api_call — call external APIs
write_file — can create or modify files
execute_sql — can modify databases
send_email — can send emails