How to Set Up Vibium MCP in Cursor

Vibium ships with a built-in MCP server, and Cursor supports MCP natively — so you can give Cursor's AI a real browser in a few lines of config. Once connected, Cursor can navigate pages, click, type, and take screenshots as part of its normal tool use, with no custom integration code.

What is MCP, and why does Cursor need it?

The Model Context Protocol (MCP) is an open standard that lets AI agents call external tools. Cursor speaks MCP, and because Vibium exposes an MCP server out of the box, Cursor can drive a browser the same way it reads your files or runs commands — no glue code in between.

This is the same built-in server you'd register with Claude Code; only the editor-side configuration differs.

How do you add Vibium to Cursor?

Cursor configures MCP servers in an mcp.json file. You can scope it to a single project (.cursor/mcp.json in the repo) or apply it globally to every project.

A server entry has the same shape MCP uses everywhere — a command plus its arguments:

{
  "mcpServers": {
    "vibium": {
      "command": "npx",
      "args": ["-y", "vibium", "mcp"]
    }
  }
}

The command + args pair above is the standard MCP server shape Vibium documents — the same npx -y vibium mcp invocation works in any MCP-aware client. The exact mcp.json location and any Cursor-specific fields depend on your Cursor version, so check Cursor's official MCP documentation for the current path before relying on it.

If you built Vibium locally instead of using npx, point the command at the binary directly:

{
  "mcpServers": {
    "vibium": {
      "command": "/path/to/vibium",
      "args": ["mcp"]
    }
  }
}

How do you confirm it's connected?

After saving mcp.json, restart Cursor so it re-discovers MCP tools — discovery happens on startup, not live. Cursor's MCP settings panel should then list vibium with its available browser tools.

Then just ask Cursor in chat:

"Open example.com, take a screenshot, and tell me the main heading."

Cursor will call the Vibium MCP tools to launch Chrome, navigate, capture the screenshot, and read the content back. Chrome downloads automatically the first time, so the initial run may take a moment longer.

Useful options

The Vibium MCP server accepts the same flags regardless of which editor launches it. Pass them in the args array:

{
  "mcpServers": {
    "vibium": {
      "command": "npx",
      "args": ["-y", "vibium", "mcp", "--headless"]
    }
  }
}

• --headless runs the browser without a visible window — handy for background tasks.
• --screenshot-dir ./screenshots changes where screenshots are saved (default: ~/Pictures/Vibium/).
• --screenshot-dir "" disables file saving and returns base64 inline only.

These flags are part of the vibium mcp command itself, so they behave the same no matter which editor launches the server.

Why this matters

Giving Cursor browser access turns an AI editor into one that can verify its own work — open the page it just built, check the rendered UI, and confirm a flow actually behaves. That feedback loop is exactly what Vibium's BiDi-first design was built for.

Next steps

• Set up Vibium MCP in Claude Code
• What is Vibium?
• Vibium vs Playwright