Connect tools to Claude Desktop
Claude Desktop can launch a local MCP server as a separate process. That process runs with your operating system permissions, not inside a safe browser sandbox. Start with one well-understood server and a test folder or restricted service account. A successful connection means tools are available; it is not an endorsement of the code or the results those tools return.
Configuration location
Config file · macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
Use Settings → Developer → Edit Config when available. Do not assume a Linux configuration path; check the client documentation for your platform. Remote servers use the app’s connector flow when supported, not the local stdio file shown here.
1. Confirm the runtime before editing
Read the server README and install the required Node.js, Python, or other runtime. Run its documented version check in a terminal. Desktop applications may not inherit your shell PATH, so an absolute executable path may be needed if the app cannot find the command.
2. Merge one server into the file
Back up the existing file, then add an entry under mcpServers. Keep existing entries and preserve valid JSON: no comments or trailing commas. Copy the command, argument array, and any required environment variables from the server’s documentation. Do not paste secrets into a shared screenshot or repository.
3. Quit and reopen the application
Fully quit Claude Desktop rather than only closing its window, then launch it again. Inspect the available tools or developer diagnostics. If the process exits immediately, compare the configured command with the README and check the application’s server logs.
4. Test with a harmless request
Ask for a read-only task against nonsensitive data. Review each requested tool action before approving it. Restrict directories and credentials at the server level, because a model instruction is not an access-control boundary.
Configuration example
{
"mcpServers": {
"playwright": {
"command": "npx",
"args": [
"@playwright/mcp@latest"
]
}
}
}This example launches a browser server via npx. Review the package and its current README before running it. Merge the named entry into your existing configuration, and do not commit secrets. Examples demonstrate the file format; they do not bypass a client’s trust or approval prompts.
Playwright MCP installation and toolsTroubleshooting
If tools do not appear, check the JSON syntax, executable path, and runtime version first. Connector availability and menu labels can change between releases; check the client documentation rather than moving remote URLs into an undocumented JSON field.
Client versions and menus change. When the installed version differs from this guide, prefer its current documentation.
