The Complete Guide to MCP Config Files — Claude Desktop, Cursor, Lovable, and More

MCP (Model Context Protocol) has exploded in adoption. Hundreds of servers exist for everything from Supabase and GitHub to Figma and Notion. But every time you want to add one to your workflow, you hit the same question: where exactly does this config go, and what format does it need to be in?

The answer depends on which client you're using. This guide covers the six most popular MCP clients in 2026 — with exact file paths, format examples, and common pitfalls for each one.

Understanding MCP Server Types First

Before diving into client-specific configs, you need to know which type of server you're configuring:

1. Claude Desktop

Claude Desktop was the first mainstream MCP client and still has the largest ecosystem. It reads from a JSON config file on your machine.

Config file locations:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Remote server example:

{
  "mcpServers": {
    "supabase": {
      "url": "https://mcp.supabase.com/sse",
      "headers": {
        "Authorization": "Bearer YOUR_ACCESS_TOKEN"
      }
    }
  }
}

STDIO server example (npm):

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

2. Cursor

Cursor uses the same mcpServers format as Claude Desktop, which means you can often copy configs directly between the two. You have two scope options:

• Global (all projects): ~/.cursor/mcp.json
• Project-scoped: .cursor/mcp.json (inside your project root)

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["-y", "@playwright/mcp@latest"],
      "env": {}
    },
    "notion": {
      "url": "https://mcp.notion.com/sse",
      "headers": {
        "Authorization": "Bearer YOUR_NOTION_TOKEN"
      }
    }
  }
}

3. Lovable

Lovable is one of the fastest-growing AI app builders in 2026, letting you build full-stack apps with natural language. MCP support in Lovable gives your projects live access to external APIs and databases during development.

In Lovable, you configure MCP servers through the project settings UI — there's no manual JSON file to edit. Navigate to:

Project Settings → Integrations → MCP Servers → Add Server

Paste the server URL and any required auth token. Lovable stores credentials securely in the project's secret vault — they're never exposed in the codebase. STDIO servers are not supported in Lovable (it's a browser-based environment); only remote HTTP/SSE servers work.

4. Cline

Cline is a powerful open-source AI coding agent (VS Code extension) with strong MCP support. It manages MCP servers through its own settings panel in VS Code, storing the config at:

• macOS/Linux: ~/Library/Application Support/Code/User/globalStorage/saoudrizwan.claude-dev/settings/cline_mcp_settings.json
• Windows: %APPDATA%\Code\User\globalStorage\saoudrizwan.claude-dev\settings\cline_mcp_settings.json

The format mirrors Claude Desktop's mcpServers schema exactly:

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

The easiest way to add servers in Cline is through the MCP Servers panel in the sidebar — click the plug icon, then "Install MCP Server". Cline can even auto-install servers from natural language descriptions.

5. Continue

Continue is an open-source AI coding assistant that works in both VS Code and JetBrains IDEs. MCP support was added in v0.9 and has become a core part of its tooling layer.

Config lives in ~/.continue/config.json (or ~/.continue/config.yaml for YAML users). The MCP section uses a mcpServers array:

{
  "mcpServers": [
    {
      "name": "github",
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "YOUR_TOKEN_HERE"
      }
    },
    {
      "name": "supabase",
      "url": "https://mcp.supabase.com/sse",
      "requestOptions": {
        "headers": {
          "Authorization": "Bearer YOUR_ACCESS_TOKEN"
        }
      }
    }
  ]
}

6. Zed

Zed is a high-performance code editor built in Rust with native MCP support. Its config uses a different key — context_servers — inside ~/.config/zed/settings.json.

STDIO server (npm):

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

Remote server (via mcp-remote bridge):

{
  "context_servers": {
    "supabase": {
      "command": {
        "path": "npx",
        "args": [
          "-y",
          "mcp-remote",
          "https://mcp.supabase.com/sse",
          "--header",
          "Authorization: Bearer YOUR_ACCESS_TOKEN"
        ]
      }
    }
  }
}

Quick Reference: Client Config Comparison

Best Practices for MCP Config Files

Never hardcode secrets

API keys and tokens in your config file sit in plaintext on your filesystem. Use environment variables wherever possible, and if you must hardcode a token (some clients require it), make sure the file is not inside a project directory that gets committed to git.

# Add to .gitignore if your config is project-scoped
.cursor/mcp.json
.continue/config.json

Use npx -y for npm packages

The -y flag tells npx to skip install confirmation prompts. Without it, your client may hang waiting for a y/n input that never comes.

Pin versions in production

Using @latest is convenient but can break if the server releases a breaking change. Once you've validated a server version works, pin it: @modelcontextprotocol/server-github@0.6.2.

Test before adding to production config

Before adding a server to your client config, use MCP Playground to verify the server is reachable, the auth token works, and the tools behave as expected. This saves you the debug loop of editing a config file, restarting your client, and discovering the URL was wrong.

Common Config Mistakes and How to Fix Them

Generate Your Config in Seconds

Writing config JSON by hand is tedious and error-prone. The MCP Config Generator handles it automatically — paste a server URL, npm package name, or PyPI package and get a ready-to-paste config for Claude Desktop, Cursor, and Zed in one click.

Related Resources

• MCP Config Generator — auto-generate configs for any server
• MCP Server Tester — verify remote servers before adding to config
• MCP Servers Directory — browse 1,000+ servers with package info
• What Is the Model Context Protocol? — protocol fundamentals
• Build a Custom MCP Server — step-by-step Node.js guide