browser-automation

When an AI agent needs to drive a real web page, a common approach is to take a screenshot and hand it to a vision model to decide where to click. That is costly and fragile. Web Bridge (browser-automation) takes a different path: it drives a real browser via Playwright and renders the page into a structured accessibility-tree snapshot with numbered reference IDs, so the agent can click, fill, and select by ref precisely without a vision model.

The plugin exposes 22 web_-prefixed tools covering navigation, interaction, observation, script execution, and multi-tab management. It supports Chromium, Firefox, and WebKit engines, and every behavioral knob such as timeouts, viewport, proxy, and persistent context can be tuned through WEB_BRIDGE_-prefixed environment variables or a JSON config file. The snapshot mechanism blanks password fields and caps node depth by default, balancing accuracy and safety.

It is a good fit for teams embedding real-browser steps into agent pipelines for automated testing, web data collection, or cross-site workflow orchestration, and for architects who prefer a single MCP protocol to manage all tools instead of maintaining separate Playwright scripts.