What is an MCP server?

An MCP server is a lightweight program that exposes tools, resources, and prompts to an AI host (like Claude Desktop, Claude Code, or Cursor) using the Model Context Protocol — an open standard created by Anthropic. The server runs locally or remotely and communicates with the AI client over stdio or HTTP/SSE transports.

How do I add an MCP server to Claude Desktop?

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%/Claude/claude_desktop_config.json (Windows). Add an entry under the mcpServers key with a command field pointing to the server executable. Restart Claude Desktop after saving.

What is the difference between stdio and SSE MCP transports?

stdio transport runs the MCP server as a child process. The host communicates via standard input/output. It's the simplest option for local servers. SSE (Server-Sent Events) transport runs the server as an HTTP server at a URL. The host connects over HTTP. SSE is needed for remote servers, shared team infrastructure, or servers that run as persistent daemons.

MCP vs function calling: what is the difference?

Function calling (OpenAI, Anthropic tool use) defines tools inside the API request and the LLM decides when to call them. MCP is a protocol layer above the API: a separate server process exposes tools, resources, and prompts to a host application. MCP tools are reusable across multiple LLMs and host apps, versioned independently, and don't need to be redefined per prompt.

MCP Claude Desktop 9 min read

MCP Server Guide 2025: How to Build, Use & Find Model Context Protocol Servers

Model Context Protocol (MCP) is the open standard that lets AI assistants like Claude talk to external tools, databases, and APIs. This guide covers everything from adding your first server to building one from scratch.

What Is Model Context Protocol (MCP)?

MCP is an open protocol published by Anthropic in late 2024. It standardizes how AI hosts (Claude Desktop, Claude Code, Cursor, Windsurf) connect to external servers that provide:

Tools — callable functions the LLM can invoke (read a file, run a query, search the web)
Resources — data sources the host can read (files, database rows, API responses)
Prompts — reusable prompt templates the user can invoke via slash commands

Think of MCP as USB-C for AI tools: one standard connector that works across hosts and servers. Anthropic, Microsoft, Google, and hundreds of open-source contributors have adopted it. Browse 300+ community servers in the Prismix MCP directory.

MCP Architecture: Client, Server & Transport

Host

The AI application: Claude Desktop, Claude Code, Cursor, Zed, Continue.dev. The host manages connections and surfaces MCP capabilities to the LLM.

MCP Client

A component inside the host that maintains a 1:1 connection to one MCP server. The host manages multiple clients (one per server).

MCP Server

A separate process or remote service that exposes tools/resources/prompts over the MCP protocol. Each server is independent and focused on one domain (filesystem, GitHub, Postgres, etc.).

Transport

stdio: server runs as a child process, communicates via stdin/stdout. Best for local servers. SSE (Server-Sent Events): server runs as an HTTP service at a URL. Best for remote/shared servers.

How to Use MCP Servers in Claude Desktop

Claude Desktop reads its MCP configuration from a JSON file. Edit it to add servers:

# macOS
~/Library/Application Support/Claude/claude_desktop_config.json

# Windows
%APPDATA%\Claude\claude_desktop_config.json

Example config adding a filesystem server and a GitHub server:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/you/projects"],
      "env": {}
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_..."
      }
    }
  }
}

Restart Claude Desktop after saving. You'll see a hammer icon in the chat UI confirming tools are available. See our curated server list for 15 ready-to-paste configs.

How to Build an MCP Server

Official SDKs exist for Python and TypeScript. A minimal Python MCP server:

pip install mcp

from mcp.server.fastmcp import FastMCP

mcp = FastMCP("My Server")

@mcp.tool()
def get_weather(city: str) -> str:
    """Return current weather for a city."""
    return f"Sunny, 22°C in {city}"   # replace with real API call

if __name__ == "__main__":
    mcp.run()   # stdio transport by default

The TypeScript equivalent:

npm install @modelcontextprotocol/sdk

import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
import { z } from "zod";

const server = new McpServer({ name: "My Server", version: "1.0.0" });

server.tool("get_weather", { city: z.string() }, async ({ city }) => {
  return { content: [{ type: "text", text: `Sunny in ${city}` }] };
});

const transport = new StdioServerTransport();
await server.connect(transport);

Popular MCP Server Categories

Category	Examples	What it enables
Filesystem	`@mcp/server-filesystem`	Read/write local files, search directories
Databases	Postgres, SQLite, Supabase MCP	Run SQL queries, inspect schemas
Dev tools	GitHub, GitLab, Jira, Linear	Create issues, review PRs, search code
Web / search	Brave Search, Fetch, Puppeteer	Browse pages, search the web, scrape data
APIs	Stripe, Twilio, HubSpot, Slack	Trigger business workflows directly from chat
Monitoring	Prismix MCP, Datadog, Sentry	Check AI service status, query alerts, view incidents

MCP vs Function Calling

Factor	MCP	Function Calling
Where defined	Separate server process	Inside API request payload
Reusability	Reuse across hosts and LLMs	Redefined per request/application
Resources + Prompts	Yes (first-class primitives)	No
Multi-host	Works with any MCP-compatible host	Provider-specific
Best for	IDE integrations, dev environments, desktop apps	Server-side API integrations, custom LLM apps

Where to Find MCP Servers

Prismix MCP directory — 300+ curated servers with install commands, GitHub stats, and community ratings
GitHub topic mcp-server — search topic:mcp-server on GitHub for community projects
Anthropic's reference servers — github.com/modelcontextprotocol/servers (official, audited)
Anthropic's blog — new server announcements from vendors (Cloudflare, Stripe, GitHub, etc.)

Track MCP Server Status

MCP servers often depend on external APIs. Check live status for Anthropic, GitHub, Stripe, and 80+ AI services from one dashboard — get instant alerts when any dependency degrades.

Browse MCP Servers →