Cloudflare Rebrands Browser Rendering to Browser Run for AI Agents

Cloudflare is doubling down on AI agents with a significant rebranding and feature rollout. The company has renamed its Browser Rendering service to Browser Run, signaling a strategic pivot to serve the burgeoning needs of AI agents that require web interaction capabilities. This move underscores the growing demand for tools that allow artificial intelligence to navigate, process, and act upon information found on the internet.

AI agents need to perform complex tasks online, from browsing websites and filling forms to extracting data and capturing screenshots. They also require oversight, with mechanisms for human intervention when automation falters. Cloudflare's Browser Run aims to provide a robust platform for these operations at scale.

The Browser for AI Agents

The rebranding from Browser Rendering to Browser Run better reflects the product's core functionality: enabling users to run full browser sessions on Cloudflare's global network. These sessions can be driven by code or AI, recorded for debugging, and allow for real-time human intervention.

New Features Enhance Agent Capabilities

Several key features are being introduced to bolster Browser Run's utility for AI agents:

• Live View: Offers real-time visibility into what an AI agent is seeing and doing, allowing for immediate detection of issues.
• Human in the Loop: Enables an agent to hand off control to a human operator when encountering unexpected roadblocks, such as login pages or complex edge cases, before resuming automation.
• Chrome DevTools Protocol (CDP) Endpoint: Exposes the CDP directly, granting AI agents granular control over browser functions and ensuring compatibility with existing CDP scripts. This direct access is crucial for advanced AI agent browser automation, potentially simplifying workflows previously managed by tools like Cursor Agents.
• MCP Client Support: Integrates with AI coding assistants like Claude Desktop and Cursor, allowing them to use Browser Run as a remote browser.
• WebMCP Support: Introduces a new browser API from Google Chrome that enables websites to declare available actions for AI agents, improving navigation reliability.
• Session Recordings: Captures complete browser sessions, including DOM changes and user interactions, for post-execution debugging.
• Higher Limits: Increases the concurrent browser limit to 120, a fourfold increase from the previous 30, enabling larger-scale operations.

These enhancements aim to address the fundamental challenges of AI agent web interaction, making them more reliable and manageable.

Under the Hood: Control and Observability

Browser Run provides multiple control mechanisms. Beyond existing support for Puppeteer and Playwright, the direct CDP endpoint offers deep browser control, enabling actions like JavaScript debugging and more efficient communication with AI models. This is particularly relevant for agents needing to analyze and interact with web pages dynamically, a core aspect of comprehensive AI agent web interaction.

For scenarios requiring less direct control, Cloudflare also offers 'Quick Actions' for simpler tasks like screenshots and PDF generation. The new WebMCP support is designed to make the web more agent-friendly by allowing sites to expose predefined tools and actions directly to AI agents.

Observability is significantly improved with Live View and Session Recordings. Live View provides a real-time window into agent activity, while Session Recordings offer a detailed historical log for post-mortem analysis. The redesigned dashboard consolidates these activities, providing a unified view of all browser sessions and quick actions.

Human Oversight and Scalability

The Human in the Loop feature is crucial for bridging the gap between AI capabilities and real-world complexity. It allows for seamless transitions between automated execution and human intervention, ensuring critical tasks can be completed even when agents encounter unforeseen issues.

Cloudflare has also addressed scalability concerns by quadrupling the concurrent browser limit to 120. This allows developers to run more AI agent tasks simultaneously, significantly boosting operational throughput.

Cloudflare's Browser Run is now available on both its Workers Free and Paid plans.