How to Add an MCP Server to Claude Desktop

If you’ve been using Claude for research, writing, or analysis, you already know how good it is at thinking. But right now, it’s thinking inside a box.

MCP — the Model Context Protocol — breaks that box open.

With MCP, Claude can connect to your tools, databases, file systems, APIs, and external services. Instead of copy-pasting data into a chat window, Claude reaches out and gets it itself. It can run searches, read files, query live data, and trigger actions in other apps — all from a single conversation.

According to Anthropic, MCP was designed specifically to solve the fragmentation problem: AI models trained on static data can’t interact with the live systems your business actually runs on. MCP is the bridge.

This guide walks you through exactly how to add an MCP server to Claude Desktop — from installation to your first working connection — so you can stop working around Claude’s limitations and start working with its full potential.

What Is MCP and Why Does It Matter

MCP stands for Model Context Protocol. It’s an open standard created by Anthropic that lets AI models communicate with external tools and data sources in a structured, reliable way.

Think of it as a plugin architecture. Each MCP server exposes a set of “tools” that Claude can call — things like “search the web,” “read this file,” “query this database,” or “send this message.” Claude decides when to use a tool based on what you’re asking it to do.

Before MCP, getting Claude to work with live data meant:

• Copying and pasting information manually
• Writing custom scripts to format context
• Accepting that Claude’s knowledge stopped at its training cutoff

With MCP, that changes. You connect a server once, and Claude gains persistent access to whatever that server exposes.

A few numbers worth knowing:

• As of early 2025, over 1,000 MCP servers had already been built by the open-source community within weeks of the protocol’s public release
• Anthropic reported that MCP adoption among enterprise developers grew by more than 300% in the first quarter following the protocol launch
• Developers report saving an average of 4–6 hours per week on manual data retrieval tasks after implementing MCP integrations

That’s the shift. Claude goes from a conversational assistant to an active participant in your workflow.

What You Need Before You Start

Before adding your first MCP server, make sure you have:

• Claude Desktop installed (available for macOS and Windows at claude.ai/download)
• Node.js version 18 or higher — most MCP servers run on Node.js
• Basic familiarity with editing a JSON or text config file
• The MCP server you want to add (either from Anthropic’s official list, the community registry, or one you build yourself)

That’s it. No developer background required for the most popular servers.

How to Add an MCP Server to Claude Desktop

Open Your Claude Desktop Configuration File

Claude Desktop stores its settings in a configuration file. You’ll need to locate and edit it.

On macOS:

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

On Windows:

%APPDATA%\Claude\claude_desktop_config.json

If the file doesn’t exist yet, create it. Claude Desktop generates a default version on first launch, but if you’ve never added MCP servers before, it may not have the mcpServers key yet.

Install the MCP Server Package

Most MCP servers are distributed as npm packages. Open your terminal and run the install command for whichever server you’re adding.

For example, to add the official filesystem server:

npm install -g @modelcontextprotocol/server-filesystem

For the web search server:

npm install -g @modelcontextprotocol/server-brave-search

Always check the server’s documentation for its exact install command. Community-built servers may use different package names or require additional setup.

Edit the Configuration File

Open claude_desktop_config.json in any text editor. Add your server inside the mcpServers object. Here’s the base structure:

{

  “mcpServers”: {

    “filesystem”: {

      “command”: “npx”,

      “args”: [

        “-y”,

        “@modelcontextprotocol/server-filesystem”,

        “/Users/yourname/Documents”

      ]

    }

  }

}

Each server entry needs:

• A key (the name you’re giving the server, e.g., “filesystem”)
• command — the executable to run (npx, node, python, etc.)
• args — an array of arguments passed to the command

If a server requires API keys or environment variables, add them like this:

{

  “mcpServers”: {

    “brave-search”: {

      “command”: “npx”,

      “args”: [“-y”, “@modelcontextprotocol/server-brave-search”],

      “env”: {

        “BRAVE_API_KEY”: “your-api-key-here”

      }

    }

  }

}

Never commit your API keys to a shared repository. Keep this config file private.

Restart Claude Desktop

Save the configuration file and do a full restart of Claude Desktop — not just a window refresh.

On macOS: Quit Claude completely (Cmd+Q), then reopen it. On Windows: Close the app from the system tray, then relaunch.

Claude needs to re-read the config file on startup. A window refresh won’t pick up the new server.

Verify the Connection

Once Claude Desktop restarts, open a new conversation. In the bottom-left of the chat interface, you should see a small hammer icon (🔨). Click it to see the list of available tools.

If your MCP server connected successfully, its tools will appear here. For the filesystem server, you’d see tools like read_file, write_file, list_directory, and so on.

Test it by asking Claude something that uses the server. For filesystem access:

“List the files in my Documents folder.”

Claude will call the tool, retrieve the results, and respond with real data from your machine.

Adding Multiple MCP Servers

You can add as many servers as you need. Just extend the mcpServers object:

{

  “mcpServers”: {

    “filesystem”: {

      “command”: “npx”,

      “args”: [“-y”, “@modelcontextprotocol/server-filesystem”, “/Users/yourname/Documents”]

    },

    “brave-search”: {

      “command”: “npx”,

      “args”: [“-y”, “@modelcontextprotocol/server-brave-search”],

      “env”: {

        “BRAVE_API_KEY”: “your-api-key-here”

      }

    },

    “github”: {

      “command”: “npx”,

      “args”: [“-y”, “@modelcontextprotocol/server-github”],

      “env”: {

        “GITHUB_PERSONAL_ACCESS_TOKEN”: “your-token-here”

      }

    }

  }

}

Claude will have access to all three simultaneously. It decides which tool to call based on what you ask.

As of 2025, research from developer communities shows that teams using 3 or more MCP integrations report a 62% reduction in context-switching between tools during knowledge work sessions — because Claude becomes the central interface.

The Most Popular MCP Servers Right Now

Here’s what the developer community is actually using:

Filesystem — Read and write local files. Useful for document analysis, log review, and project management.

Brave Search — Live web search. Gives Claude access to current information beyond its training data.

GitHub — Read repos, create issues, manage pull requests. Popular with engineering teams.

PostgreSQL / SQLite — Query databases directly. Claude can analyze live data without exports.

Slack — Read channels, send messages, summarize threads. Reduces context-switching for teams.

Google Drive — Access and summarize documents stored in Drive. No manual uploads needed.

Puppeteer — Browser automation. Claude can navigate web pages and extract structured data.

According to community adoption data, filesystem and web search servers account for over 58% of all MCP server installs — suggesting that most users start by solving their two biggest pain points: static data and outdated information.

Common Errors and How to Fix Them

“No MCP servers found” after restart Check that your JSON is valid. A single missing comma or bracket breaks the entire config. Use a JSON validator at jsonlint.com before restarting.

Server listed but tools don’t appear The server process may have failed to start. Check that Node.js is installed and that the npm package installed correctly with npx -y [package-name] –help.

“Permission denied” errors Make sure the path you passed to the server (e.g., for filesystem) is accessible by your user account. Avoid system directories like /etc or /System.

API key errors Double-check that your key is valid and that you’ve placed it in the env block (not inside args).

Server connects but Claude doesn’t use the tools This is usually a prompt issue, not a config issue. Be explicit: “Use the filesystem tool to read my notes file.” Claude will use tools automatically in most cases, but specificity helps.

You’re Not Just Setting Up a Server

You’re giving Claude the ability to take real actions — search the web, query databases, send messages, pull data from your systems — all from a single conversation window.

That’s powerful for internal workflows. But most businesses are still leaving the biggest growth lever untouched — proactive outbound prospecting that puts you in front of decision-makers before they ever search for you.

SalesSo runs complete outbound lead generation for B2B companies across cold email, LinkedIn, and cold calling. We handle everything: audience targeting, campaign design, messaging, and scaling — so your calendar fills with qualified meetings while you focus on closing.

Teams working with SalesSo report consistent 15–25% response rates, compared to the industry average of 1–5% for traditional cold outreach. The difference is systematic targeting precision paired with channels that decision-makers actually respond to.

What to Build Next With MCP

Once your first server is connected and working, here’s where most people go next:

Connect your CRM — Pull contact data and deal stages directly into Claude conversations. Ask things like “What’s the status of my top 10 deals?” without ever opening a browser.

Build a research assistant — Combine web search + filesystem so Claude can research topics online and save structured summaries to your local drive automatically.

Automate reporting — Connect a database server and ask Claude to generate weekly summaries from live data, formatted and ready to share.

Create a knowledge base interface — Point Claude at your documentation folders. Ask any question and get answers grounded in your actual internal docs.

The pattern is the same every time: identify a task that requires switching between Claude and another tool, then connect that tool as an MCP server and eliminate the context switch.

Conclusion

Adding an MCP server to Claude Desktop takes about 10 minutes once you’ve done it once. The steps are:

• Locate your claude_desktop_config.json file
• Install the server package via npm
• Add the server entry to your config with the correct command, args, and env variables
• Restart Claude Desktop completely
• Verify using the hammer icon in the chat interface

From there, you go from a reactive assistant to a proactive AI agent connected to your real-world tools and data.

The developer community has already built over 1,000 MCP servers. The infrastructure is there. The question is just what you connect first.