The Ultimate Guide to Installing MCP Servers in Claude Desktop (2026 Edition)

⚡ TL;DR: Quick Answer

To install an MCP server in Claude Desktop, open your claude_desktop_config.json file (located in ~/Library/Application Support/Claude/ on macOS or %APPDATA%\Claude\ on Windows). Add a new object under the "mcpServers" key specifying the server's name, the execution command (e.g., npx or python), and the required args. Restart Claude Desktop, and the new tool will appear in the 🔌 (Tools) menu. For a curated list of verified servers, visit the Lord of Claude MCP Directory.

---

🧠 What is the Model Context Protocol (MCP)?

The Model Context Protocol (MCP) is an open standard developed by Anthropic that allows AI assistants (like Claude Desktop, Cursor, and Windsurf) to securely connect to external data sources, APIs, and local tools.

Before MCP, AI models were isolated in a "sandbox." With MCP, Claude can read your local files, query a PostgreSQL database, manage GitHub pull requests, or scrape live web data—all through standardized, secure server connections. Mastering MCP is the foundational skill for building agentic AI workflows.

---

🛠️ Prerequisites: What You Need Before Starting

Before modifying your Claude configuration, ensure you have the following installed on your system:

1. Claude Desktop App: Updated to the latest version (MCP support is built into recent releases).
2. Node.js (v18 or higher): Required for running JavaScript/TypeScript-based MCP servers via npx. Download Node.js.
3. Python (v3.10 or higher): Required for Python-based MCP servers. Ensure pip is installed.
4. Basic Terminal Knowledge: You will need to run simple commands to verify installations.

💡 Pro Tip: Always verify your installations by running node -v and python --version in your terminal before proceeding.

---

📝 Step-by-Step: How to Install an MCP Server in Claude Desktop

Follow this exact process to connect your first MCP server. We will use the official Filesystem MCP Server as our example, which allows Claude to safely read and write to a specific folder on your computer.

Step 1: Locate Your Claude Desktop Configuration File

The configuration file is named claude_desktop_config.json. Its location depends on your operating system:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json (Usually C:\Users\<YourUsername>\AppData\Roaming\Claude\)
• Linux: ~/.config/Claude/claude_desktop_config.json

If the file does not exist, create it manually as a blank JSON file.

Step 2: Edit the JSON Configuration

Open the file in a text editor (like VS Code or Notepad). You will add your server under the "mcpServers" object.

Copy and paste the following configuration for the Filesystem server:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "/Users/yourusername/Documents/claude-workspace"
      ]
    }
  }
}

⚠️ Critical Security Note: Replace /Users/yourusername/Documents/claude-workspace with an actual, existing directory on your machine. Never grant MCP servers access to your entire root drive (/ or C:\). Restrict access only to dedicated project folders.

Step 3: Save and Restart Claude Desktop

1. Save the claude_desktop_config.json file.
2. Completely quit the Claude Desktop application (Cmd+Q on Mac, or right-click > Exit on Windows).
3. Reopen Claude Desktop.

Step 4: Verify the Installation

Start a new chat in Claude. Click the 🔌 (Tools) icon in the bottom right of the prompt box. You should now see "filesystem" listed as an available tool.

Test it with this prompt:

"List the files in my claude-workspace directory and summarize the contents of the most recent text file."

---

🏆 Top 3 Beginner-Friendly MCP Servers to Try First

Ready to expand Claude's capabilities? Here are the most popular, highly-rated MCP servers you can add today. (Find more verified options in the Lord of Claude MCP Directory)

1. GitHub MCP Server (@modelcontextprotocol/server-github)

• What it does: Allows Claude to read repositories, create issues, review pull requests, and search code.
• Best for: Software engineers automating code reviews or managing project backlogs.
• Config Snippet:

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

2. PostgreSQL MCP Server (@modelcontextprotocol/server-postgres)

• What it does: Connects Claude directly to your database to run read-only queries, analyze schemas, and generate reports.
• Best for: Data analysts and backend developers.
• Config Snippet:

"postgres": {
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-postgres", "postgresql://user:password@localhost:5432/mydb"]
}

3. Brave Search MCP Server (@modelcontextprotocol/server-brave-search)

• What it does: Gives Claude real-time web search capabilities, bypassing its static knowledge cutoff.
• Best for: Researchers, marketers, and anyone needing up-to-the-minute information.
• Config Snippet:

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

---

🔧 Troubleshooting Common MCP Errors

Even with perfect syntax, MCP setups can fail. Here are the most common errors and how to fix them:

---

🛡️ Advanced Best Practices for MCP Security

As you integrate more tools, security becomes paramount. Follow these rules to keep your system safe:

1. Use Environment Variables for Secrets: Never hardcode API keys or database passwords directly in the JSON file if you plan to share it. Use your OS environment variables and reference them in the "env" block.
2. Principle of Least Privilege: When creating GitHub PATs or database users for MCP, grant only the permissions required (e.g., read-only access for a database analyzer).
3. Isolate Workspaces: Create a dedicated folder (e.g., ~/claude-mcp-sandbox/) for the Filesystem server. Do not point it to ~/Documents or C:\Users\Name\.
4. Audit Regularly: Review your claude_desktop_config.json monthly. Remove servers you no longer use to reduce your attack surface.

---

❓ Frequently Asked Questions (FAQ)

Q: Can I use MCP servers in the Claude web interface (claude.ai)?
A: No. Currently, MCP server execution requires a local runtime environment, which is only supported in the Claude Desktop application and local IDEs like Cursor or Windsurf.

Q: Are MCP servers free to use?
A: The Model Context Protocol itself and most community servers (like those listed in the Lord of Claude directory) are open-source and free. However, some servers may require your own API keys (e.g., Brave Search, PostgreSQL hosting), which may have associated costs.

Q: Will adding too many MCP servers slow down Claude Desktop?
A: MCP servers are designed to be lazy-loaded. They only consume resources when Claude actively calls a tool during a conversation. However, having 10+ servers may slightly increase the initial app startup time.

Q: How do I find new, verified MCP servers?
A: The safest way to discover new tools is through curated, community-vetted directories. Visit the Lord of Claude MCP Server Directory to browse hundreds of verified, starred, and categorized MCP servers updated daily.

---

🚀 Conclusion: Unlock Claude's True Potential

The Model Context Protocol is fundamentally changing how we interact with AI. By learning how to install MCP servers in Claude Desktop, you transform Claude from a passive chatbot into an active, context-aware agent capable of executing real-world tasks.

Start with the Filesystem server, experiment with GitHub or Database connectors, and gradually build your ultimate AI workflow.

🔗 Ready to explore more? Browse the Lord of Claude Directory to discover trending Claude Skills, verified MCP servers, and the latest AI infrastructure news. Have a custom server? Submit it to our directory to get featured in front of thousands of AI developers.

---