is an (by Anthropic, Nov 2024) that gives AI models a universal, standardized way to connect to , , and — no matter which AI app or model is using them.
概览
is an (by Anthropic, Nov 2024) that gives AI models a universal, standardized way to connect to , , and — no matter which AI app or model is using them. Think of it as — one standard connector for everything. Before MCP, every AI application had to build its own integrations: With MCP, you build an integration once as an MCP Server, and any AI app can use it:
README
MCP — Model Context Protocol
MCP is an open standard (by Anthropic, Nov 2024) that gives AI models a universal, standardized way to connect to tools, data, and services — no matter which AI app or model is using them.
Think of it as USB-C for AI — one standard connector for everything.
The Problem MCP Solves
Before MCP, every AI application had to build its own integrations:
App A → custom code → Database
App A → custom code → GitHub
App A → custom code → Slack
App B → different custom code → Database (built twice!)
App B → different custom code → GitHub (built twice!)
With MCP, you build an integration once as an MCP Server, and any AI app can use it:
MCP Server: Database ←── App A (Claude Desktop)
←── App B (Cursor)
←── App C (your custom app)
Folder Structure
mcp/
├── README.md
├── concepts/
│ ├── 01-what-is-mcp.md ← The big idea
│ ├── 02-architecture.md ← Host / Client / Server
│ ├── 03-tools.md ← Callable functions (like function calling)
│ ├── 04-resources.md ← Readable data (like RAG)
│ ├── 05-prompts.md ← Reusable prompt templates
│ ├── 06-transports.md ← stdio vs HTTP/SSE
│ ├── 07-mcp-vs-function-calling.md ← Key differences
│ └── 08-best-use-cases.md ← Real-world applications
│
└── implementation/
├── requirements.txt
├── 01_minimal_server.py ← Simplest MCP server (2 tools)
├── 02_tools_server.py ← Server with 6 real tools
├── 03_resources_server.py ← Server exposing files as resources
├── 04_full_server.py ← Tools + Resources + Prompts
└── 05_client.py ← MCP client connecting to a server
Three Things an MCP Server Can Expose
| Primitive | What it is | Equivalent |
|---|---|---|
| Tools | Functions the LLM can call | Function Calling |
| Resources | Data/files the LLM can read | RAG / Context |
| Prompts | Reusable prompt templates | System prompts |
Quick Start
cd implementation
pip install -r requirements.txt
# Run the full server
python 04_full_server.py
# In another terminal, connect the client
python 05_client.py
Read Concepts in Order
| # | Topic |
|---|---|
| 1 | What is MCP and why it exists |
| 2 | Architecture: Host, Client, Server |
| 3 | Tools — the callable functions |
| 4 | Resources — the readable data |
| 5 | Prompts — reusable templates |
| 6 | Transports: stdio vs HTTP/SSE |
| 7 | MCP vs Function Calling |
| 8 | Best use cases |
安装
This server does not publish a one-line install command.
Open the repository installation guide 新手？先看 mcpskill.org 教程 →