TL;DR
Vibium is a new browser automation tool that bundles a small Go binary with a JavaScript client to let AI agents and developers control Chrome with minimal setup. The project exposes a BiDi WebSocket and an MCP server so agents like Claude Code can drive a browser out of the box; npm publishing and cross‑platform support are already in place.
What happened
Vibium, released as an open repository and published to npm, provides a compact Go binary called Clicker plus a JS/TS client to simplify browser automation for both AI agents and human developers. The Clicker binary (around 10MB) manages Chrome lifecycle, acts as a WebDriver BiDi proxy over WebSocket (:9515) and exposes an MCP server on stdio so MCP‑compatible agents can send commands without manual browser setup. The JavaScript package offers both synchronous and asynchronous APIs for common actions — navigate, find, click, type, screenshot and quit — and installs a platform‑appropriate Clicker and Chrome for Testing into a local cache unless the user opts out. The repo lists platform support across Linux, macOS (Intel and Apple Silicon) and Windows, tracks daily progress updates, is released under Apache 2.0, and publishes a V2 roadmap that includes additional clients and recording features.
Why it matters
- Reduces setup friction for LLM agents to interact with real web pages by packaging browser lifecycle, BiDi proxy and MCP server into one binary.
- Provides developer‑facing JS APIs (sync and async) so human engineers can automate browsing tasks without managing browser drivers manually.
- Cross‑platform packaging and automatic Chrome download lower barriers for test automation and agent-based web tasks on common OSes.
- Direct integration with MCP clients (example: Claude Code) enables immediate agent-driven browsing with a single install command.
Key facts
- Clicker is a single Go binary (~10MB) that handles browser management, BiDi proxying and an MCP server.
- Vibium exposes a WebSocket WebDriver BiDi endpoint on port :9515 and an MCP server over stdio for agents.
- A JavaScript/TypeScript client is available via npm (package name: vibium) offering both sync and async APIs.
- Agent integration example: 'claude mcp add vibium — npx -y vibium' installs Vibium for Claude Code and triggers automatic Chrome download.
- Common tool actions implemented include browser_launch, browser_navigate, browser_find, browser_click, browser_type, browser_screenshot and browser_quit.
- Default behavior installs Chrome for Testing and chromedriver into OS‑specific caches (Linux, macOS, Windows); download can be skipped via VIBIUM_SKIP_BROWSER_DOWNLOAD=1.
- Platform matrix in the repo lists support for Linux x64, macOS x64 and arm64, and Windows x64.
- Project is open source under the Apache 2.0 license and was published to npm on 2025-12-22 according to the changelog.
What to watch next
- Planned V2 clients for Python and Java listed in the project's roadmap.
- Roadmap items include Cortex (memory/navigation layer), Retina (recording extension) and video recording.
- AI‑powered locators are listed as a planned feature for future releases.
Quick glossary
- WebDriver BiDi: A bidirectional WebDriver protocol that uses WebSocket to send commands to and receive events from a browser.
- MCP server: An agent communication interface exposed by Vibium; in this project it refers to a message/control protocol endpoint available over stdio for LLM agents.
- Clicker (Vibium Clicker): The lightweight Go binary included with Vibium that manages browser lifecycle, BiDi proxying and the MCP server.
- npm: Node package manager and registry used to distribute JavaScript packages, including the Vibium client library.
Reader FAQ
Is Vibium open source?
Yes; the project is released under the Apache 2.0 license.
How do I enable Vibium for Claude Code?
The repository shows a one‑line example: 'claude mcp add vibium — npx -y vibium' which installs Vibium for Claude Code.
Does Vibium automatically download a browser?
Yes — during setup it downloads Chrome for Testing by default and caches it per OS; this can be skipped with an environment variable.
Are Python and Java clients available now?
The roadmap lists Python and Java clients as planned items, but full clients are not confirmed as released in the source.
Vibium Browser automation without the drama. Vibium is browser automation infrastructure built for AI agents. A single binary handles browser lifecycle, WebDriver BiDi protocol, and exposes an MCP server —…
Sources
- Show HN: Vibium – Browser automation for AI and humans, by Selenium's creator
- Vibium: Browser Automation for the AI Era — What's …
- Vibium Architecture Explained: AI-Native Browser Automation …
- Vibium: The AI-Native Successor to Selenium
Related posts
- Minimalist browser-based text editor that compresses your document into the URL
- Jolla launches a new phone with Sailfish 5 to counter iOS–Android sameness
- DJ Garman drops the ball instead of the bass in AWS re:Invent keynote