← AI Terminology
MCP Server
An MCP server (Model Context Protocol server) is a lightweight service that implements Anthropic's open Model Context Protocol — exposing tools, resources, and prompts to AI assistants via a standardised JSON-RPC interface, enabling any compliant LLM client to call the server's capabilities.
MCP is the emerging standard for connecting AI agents to external tools and data sources.
MCP is the emerging standard for connecting AI agents to external tools and data sources.
Why It Matters in AI
Before MCP, every LLM application built its own bespoke integration for tools (web search, code execution, databases). MCP standardises this: an MCP server for Postgres can be plugged into Claude, Cursor, or any MCP-compliant client without custom glue code. It is analogous to what USB did for device connectivity — one interface, any device. As AI agents proliferate, MCP is rapidly becoming the de-facto protocol for the AI tool ecosystem.
Key Points
| Aspect | Description |
|---|---|
| SDK | @modelcontextprotocol/sdk (TypeScript), mcp (Python) — official Anthropic SDKs |
| Protocol | JSON-RPC 2.0 over stdio or HTTP/SSE — simple, language-agnostic, easily implementable |
| Ecosystem | MCP servers for: filesystem, Postgres, GitHub, Slack, Brave Search, Puppeteer, and hundreds more |
| Primitives | Tools (callable functions), Resources (readable data), Prompts (reusable templates) |
| Transports | stdio (local subprocess), HTTP with Server-Sent Events (remote/cloud) |
| Client support | Claude Desktop, Cursor, Cline, VS Code Copilot, Windsurf — all support MCP natively |
Simple Analogy
A USB hub for AI tools: any USB-compatible device plugs into any USB hub; any MCP server plugs into any MCP-compatible AI client. The protocol defines the socket — you build the device, the hub handles the rest.
Common Usage Examples
npx @modelcontextprotocol/server-filesystem /Users/me/projects— filesystem MCP servernpx @modelcontextprotocol/server-postgres postgresql://localhost/mydb— Postgres MCP server- Python MCP server:
@server.tool(); async def search_db(query: str) -> str: ... claude_desktop_config.json:{"mcpServers": {"filesystem": {"command": "npx", "args": [...]}}}— client config- Claude Code uses MCP to expose tools to Claude for filesystem access and shell execution
Summary
In short: An MCP server implements Anthropic's open protocol for connecting AI assistants to external tools and data — the emerging standard that enables any LLM client to use any MCP-compatible tool without custom integration code.