VisFlow Enters Public Beta: Giving General-Purpose Agents Real Browser Execution
From planning to operating the browser
General-purpose agents are increasingly good at understanding goals and planning work, but many real tasks still end at the browser boundary. Internal systems require sign-in, pages change dynamically, and important workflows depend on visual controls rather than an API.
VisFlow is now in public beta to close that gap. Through a remote MCP service, agents in Codex, Claude Code, Cursor, WisCode, and other compatible clients can invoke browser tasks in the user’s existing Chrome session.
A visual execution loop
VisFlow reads the interface people actually see, acts on it, and verifies the resulting state. It does not depend solely on fixed DOM selectors or XPath rules. Users can also demonstrate a workflow so that the agent can learn and reuse a business process instead of rebuilding it from scratch each time.
Tasks can continue in the background while the user handles other work. Execution remains visible and controllable, with pause, takeover, and confirmation boundaries for sensitive actions.
Getting started
Install the VisFlow browser extension, sign in, and connect a supported agent client to the VisFlow MCP service. Once connected, describe the browser task in natural language and monitor its progress from the extension.
The public beta is intended to help teams test browser automation against real internal and public web systems while preserving the safeguards required for enterprise use.