VLearnVibium

Vibium MCP: A Complete Overview

Vibium MCP is a built-in Model Context Protocol server that gives AI agents a real browser — navigate, click, type, and screenshot with no glue code.

By Pramod Dutta··4 min read·Verified with Vibium 26.2
▶ Animated overview · made with Remotion

Vibium MCP is the built-in Model Context Protocol server that turns Vibium into a browser tool any AI agent can call. Vibium is an AI-native browser automation engine built on WebDriver BiDi, shipped as a single Go binary that auto-downloads Chrome for Testing — created by Jason Huggins, co-creator of Selenium and Appium. Because the MCP server lives inside that binary, you do not wire up a separate integration: you run npx -y vibium mcp, register it with your assistant, and the agent can navigate, click, type, scroll, and screenshot a real browser through standard MCP tool calls. It exposes around 22 documented tools (such as browser_launch, browser_navigate, browser_click, browser_type, and browser_screenshot) and works with any MCP-aware client — Claude Code, Claude Desktop, Cursor, Gemini CLI, and Windsurf. This page explains what the server does, the tools it exposes, and how to connect it.

What is MCP, and how does Vibium use it?

The Model Context Protocol (MCP) is an open standard that lets AI agents call external tools through a uniform interface. Instead of every tool needing a custom plugin, an MCP server advertises its tools and their schemas at startup, and the agent calls them like any other function.

Vibium implements an MCP server directly in its Go binary. When your assistant launches vibium mcp, it performs a discovery handshake — initialize to exchange capabilities, then tools/list to learn every browser tool and its parameters. From then on, plain-English requests like "open the page and read the heading" become real browser actions. Because Vibium is BiDi-first, those actions run against a standards-based protocol rather than a proprietary one.

Which browser tools does Vibium MCP expose?

The server exposes the full Vibium action set as MCP tools — roughly 22 in current builds. The most-used ones map directly to everyday browser work:

ToolWhat it does
browser_launchStart a browser session (visible by default)
browser_navigateGo to a URL
browser_find / browser_find_allLocate one or all matching elements
browser_clickClick an element by selector
browser_typeType text into a field
browser_get_text / browser_get_htmlRead page or element content
browser_screenshotCapture the page (learn more)
browser_evaluateRun JavaScript in the page
browser_waitWait for an element state
browser_quitClose the session

There are also tools for hovering, scrolling, key presses, dropdown selection, and tab management. As with the find element action in the client API, Vibium auto-waits for elements to be actionable, so the agent does not have to manage timing manually.

How do you connect Vibium MCP to an assistant?

Every MCP-aware client uses the same server commandnpx -y vibium mcp — and differs only in how you register it. For CLI tools, one command does it:

# Claude Code
claude mcp add vibium -- npx -y vibium mcp
 
# Gemini CLI
gemini mcp add vibium npx -y vibium mcp

Editors that read a JSON config file (Cursor, Windsurf, Claude Desktop) use the standard MCP server shape instead:

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

Detailed, client-specific guides cover the exact paths and steps: Claude Code, Claude Desktop, Cursor, Gemini CLI, and Windsurf. Tool discovery happens on startup, so restart the assistant after registering.

What options can you pass the server?

The vibium mcp command accepts flags in its args, and they behave the same regardless of which client launches them:

npx -y vibium mcp --headless
  • --headless runs Chrome without a visible window — useful for background or CI-style tasks.
  • --screenshot-dir ./screenshots changes where captures are written (default: ~/Pictures/Vibium/).
  • --screenshot-dir "" disables file saving and returns base64 inline only.

To confirm the server itself works before debugging a client, run it directly — npx -y vibium mcp should start and wait for input.

Why does a built-in MCP server matter?

Bundling MCP into the binary removes the usual integration tax. Most automation stacks need a separate bridge between the AI layer and the browser layer; Vibium ships both in one download, so an agent gains a real browser the moment the server is registered. That lets a coding assistant verify its own work — open the page it just built, read the rendered UI, and confirm a flow actually behaves — which is exactly what Vibium's design targets.

Note: Vibium is Jason Huggins' open-source project (vibium.com, github.com/VibiumDev/vibium), free via pip install vibium or npm install vibium. learnvibium.com is an independent learning hub and did not create Vibium.

Next steps

Frequently asked questions

What is Vibium MCP?

Vibium MCP is the Model Context Protocol server that ships inside the Vibium binary. It exposes browser actions — launch, navigate, click, type, screenshot — as MCP tools, so AI assistants like Claude Code, Cursor, and Gemini CLI can drive a real Chrome browser through normal tool calls.

Do I need to install anything extra to use Vibium MCP?

No. The MCP server is built into Vibium itself, started with the 'vibium mcp' command. You can run it via 'npx -y vibium mcp' with no separate package. Chrome for Testing downloads automatically the first time a browser is launched.

Which AI tools work with Vibium MCP?

Any MCP-aware client works, because Vibium speaks the open Model Context Protocol standard. Documented setups exist for Claude Code, Claude Desktop, Cursor, Gemini CLI, and Windsurf. The same 'npx -y vibium mcp' command is used everywhere; only the registration step differs.

Vibium is created by Jason Huggins. This is an independent tutorial — see the official Vibium site and GitHub repo for canonical docs.

Related guides