Browser Run adds WebMCP support for AI agent automation
Key Points
- WebMCP enables AI agents to call website functions directly with typed parameters
- Chrome beta features available through experimental browser pool with lab=true parameter
- Integrates with CDP endpoint for MCP client connections and AI agent automation
Summary
Browser Run now supports WebMCP (Web Model Context Protocol), a new browser API from the Google Chrome team that enables AI agents to discover and call website functions directly instead of relying on slow screenshot-analyze-click loops.
Key Points
- Structured tool discovery - Use
navigator.modelContextTesting.listTools()to discover available actions on WebMCP-enabled websites - Direct tool execution - Call
navigator.modelContextTesting.executeTool()with typed parameters for faster, more reliable automation - Human-in-the-loop support - Some tools can pause for user confirmation before completing sensitive actions
- Chrome beta access - WebMCP requires Chrome beta features; use
lab=trueparameter in/devtools/browserrequests to access experimental browser instances - CDP integration - Works with the recently launched CDP endpoint, allowing MCP clients to connect and enable AI agents to interact with websites directly
Benefits
Replaces unreliable screenshot-analyze-click loops with typed function calls, making browser automation faster, more reliable, and less fragile for AI agents.