🔌 What is MCP
MCP (Model Context Protocol) is a standard protocol open sourced by Anthropic for connecting AI assistants with external data sources and tools. It solves the problem of isolation of LLM applications and data, allowing AI to securely and controllably access external systems.
MCP uses a client-server architecture. AI assistants (such as Claude) act as clients, and MCP Server provides tools and data access capabilities. A unified protocol replaces fragmented integration solutions.
What MCP can do:
- Tool call : Let AI perform external operations (such as sending emails, creating Issues)
- Data Access : Let AI read external data (such as database, API, file)
- Contextual enhancement : Provide real-time external information to AI
✨ Why use MCP
MCP allows AI to change from "only talking" to "able to do things", and from a closed system to an open platform.
Main advantages:
- Unified interface : One protocol to connect all external services without writing custom code for each service
- Safe and controllable : clear permission boundaries, AI can only access authorized functions
- Rich ecology : 400+ official and community servers, covering mainstream services
- Easy to expand : Supports multiple programming language SDKs and can customize the server
📦 Installation configuration
Claude Code configuration method
Created in the project root directory
.mcp.json
File:
{
"mcpServers": {
"filesystem": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/allowed/dir"]
},
"github": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-github"],
"env": {
"GITHUB_PERSONAL_ACCESS_TOKEN": "your-token-here"
}
},
"memory": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-memory"]
}
}
}
Claude Desktop configuration method
Edit the Claude Desktop configuration file:
# macOS ~/Library/Application Support/Claude/claude_desktop_config.json # Windows %APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"brave-search": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-brave-search"],
"env": {
"BRAVE_API_KEY": "your-api-key"
}
},
"slack": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-slack"],
"env": {
"SLACK_BOT_TOKEN": "xoxb-your-token",
"SLACK_TEAM_ID": "T0123456789"
}
}
}
}
Do not submit configuration files containing API keys to version control. It is recommended to use environment variables or key management tools.
⭐ Recommended server list
Official reference server
📁 Filesystem Official
Secure file operations, supporting configurable directory access control
🔗 Fetch Official
Web content acquisition and conversion, optimized LLM usage
🧠 Memory Official
Persistent memory system based on knowledge graph
📂 Git Official
Read, search, and operate Git repositories
🕐 Time Official
Time and time zone conversion function
🤔 Sequential Thinking Official
Dynamic problem solving through thought sequences
Development tools
| Server | Function | Installation command |
|---|---|---|
| GitHub | GitHub warehouse operation, Issue, PR management | @modelcontextprotocol/server-github |
| GitLab | GitLab project management, supports OAuth | @modelcontextprotocol/server-gitlab |
| Linear | Linear Issue Search, Create, Update | @modelcontextprotocol/server-linear |
| Sentry | Error tracking and analysis | @modelcontextprotocol/server-sentry |
Database class
| Server | Function |
|---|---|
| PostgreSQL | Read-only database access and schema checking |
| SQLite | Local SQLite database operations |
| MongoDB | MongoDB database operations |
| Redis | Redis key-value storage interaction |
| Pinecone / Qdrant | Vector database semantic search |
Cloud services
☁️ AWS
AWS service integration, following best practices
🔷 Azure
Azure Storage、Cosmos DB、CLI
🌩️ Cloudflare
Workers, KV, R2, D1 Resource Management
▲ Vercel
Log access, document search, project management
Search and Data Class
🔍 Brave Search
Web and local search, requires API Key
🔎 Exa
AI dedicated search engine
🗺️ Google Maps
Location services, directions, location details
🐦 Twitter/X
Tweet search and publishing
Browser Automation
🎭 Playwright
Browser Automation Testing and Operations
🤖 Puppeteer
Browser Automation and Web Scraping
🚀 How to use
After configuration is complete, Claude will automatically identify available MCP tools. You can:
Directly describe the task
# Claude 会自动调用 GitHub MCP "帮我创建一个 Issue,标题是 'Bug: 登录页面样式错乱'" # Claude 会自动调用 Filesystem MCP "读取 src/config.ts 文件内容" # Claude 会自动调用 Brave Search MCP "搜索一下 Next.js 15 的新特性"
View available tools
Available in Claude Code
/mcp
command to view configured MCP servers and the tools they provide.
Install MCP servers on demand to avoid configuring too many infrequently used services. Each server takes up startup time and resources.
📚 Related Resources
- MCP Servers official warehouse - 400+ official and community servers
- MCP official document - Protocol specification and usage guide
- MCP Server Registry - Server Registry and Search
- TypeScript SDK - Developing a custom server
- Python SDK - Developing a custom server