How to Set Up an MCP Server with Claude: Step-by-Step Tutorial

What Is an MCP Server?

Model Context Protocol (MCP) is an open standard created by Anthropic that lets AI assistants connect to external tools and data sources. An MCP server is a small program that exposes a set of tools — functions the AI can call — over a standardized interface.

When you ask Claude to "read this file" or "search the web," it's calling an MCP tool behind the scenes. The server handles the actual work; Claude handles the reasoning about what to do with the results.

MCP vs Claude Skills: What's the Difference?

Before diving in, it's worth clarifying the terminology.

MCP Servers are persistent processes that run alongside your AI client. They expose tools via the Model Context Protocol and require a running server process.

Claude Skills (SKILL.md files) are lightweight instruction files that guide Claude's behavior using existing CLI tools — no server process required. They're simpler to install and share.

Many entries on DiscoverAISkills are technically Claude skills rather than MCP servers, but the terms are often used interchangeably in the community. Both extend what Claude can do. Popular examples that blur the line include Obsidian (75,000+ installs), Notion (73,000+ installs), and Slack (36,000+ installs).

Prerequisites

Before setting up an MCP server, you need:

Claude Desktop or Claude Code installed
Node.js 18+ (check with node --version)
Basic comfort with the terminal

Method 1: Claude Desktop (GUI)

Claude Desktop has a built-in MCP configuration interface.

Step 1: Open Claude Desktop and go to Settings → Developer → MCP Servers.

Step 2: Click Add Server.

Step 3: Fill in the server details. For the official filesystem server:

{
  "name": "filesystem",
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/yourname/Documents"]
}

Step 4: Click Save and restart Claude Desktop.

Step 5: Test it. Ask Claude: "List the files in my Documents folder." It should call the filesystem tool and return results.

Method 2: Claude Code (Terminal)

Claude Code reads MCP configuration from ~/.claude/mcp.json.

Step 1: Create or edit the config file:

nano ~/.claude/mcp.json

Step 2: Add your server configuration:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/yourname"]
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "your-token-here"
      }
    }
  }
}

Step 3: Restart Claude Code. The servers start automatically.

Setting Up the GitHub MCP Server

The GitHub skill is the most popular MCP integration with 148,000+ installs. Here's the full setup:

Step 1: Create a GitHub Personal Access Token at github.com/settings/tokens. Give it repo and read:org scopes.

Step 2: Add to your MCP config:

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_your_token_here"
      }
    }
  }
}

Step 3: Test it. Ask Claude: "What are the open pull requests in my repo?"

Setting Up Web Search

For real-time web search, you have several options — each with different strengths:

Brave Search (48,000+ installs) — privacy-focused, free tier available:

{
  "mcpServers": {
    "brave-search": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-brave-search"],
      "env": { "BRAVE_API_KEY": "your-key-here" }
    }
  }
}

Tavily Search (73,000+ installs) — AI-optimized results, better for research tasks:

{
  "mcpServers": {
    "tavily": {
      "command": "npx",
      "args": ["-y", "tavily-mcp"],
      "env": { "TAVILY_API_KEY": "your-key-here" }
    }
  }
}

Multi Search Engine (106,000+ installs) — searches multiple engines simultaneously, no API key needed for basic use.

Setting Up Productivity Integrations

Once you have the basics working, these are the most popular next steps:

GOG — Google Workspace (147,000+ installs) — Gmail, Calendar, and Drive in one integration:

{
  "mcpServers": {
    "google-workspace": {
      "command": "npx",
      "args": ["-y", "@gog/mcp-server"],
      "env": { "GOOGLE_CLIENT_ID": "...", "GOOGLE_CLIENT_SECRET": "..." }
    }
  }
}

Notion (73,000+ installs) — read and write your Notion workspace.

Slack (36,000+ installs) — send messages and read channels.

Trello (32,000+ installs) — manage boards and cards.

Apple Notes (27,000+ installs) — access your Apple Notes on macOS.

Apple Reminders (22,000+ installs) — create and manage reminders.

Combining Multiple MCP Servers

The real power comes from combining servers. A typical developer setup might include:

{
  "mcpServers": {
    "filesystem": { "...": "..." },
    "github": { "...": "..." },
    "brave-search": { "...": "..." },
    "notion": { "...": "..." }
  }
}

With this setup, Claude can read your local files, manage GitHub, search the web, and update your Notion docs — all in a single conversation. Add Agent Browser (72,000+ installs) for browser automation and Nano PDF (84,000+ installs) for document reading, and you have a genuinely powerful research and development assistant.

Troubleshooting

Server fails to start

Check that Node.js is installed and up to date. Run npx -y @modelcontextprotocol/server-filesystem --help to test the package directly.

Claude doesn't use the tools

Make sure you restarted Claude after editing the config. Also check that the server name in your config matches what Claude expects.

Authentication errors

Double-check your API keys and tokens. Most MCP servers print clear error messages — check your terminal output.

Finding More MCP Servers

DiscoverAISkills has 36,000+ skills and MCP servers across every category. Browse by platform to find MCP-specific integrations, explore collections for curated lists, or check the full rankings to see what the community is actually using.