Experimental WebMCP Integration in mcp-handler
On September 18, 2026, Vercel announced the release of [email protected], adding experimental support for WebMCP—a proposed web standard designed to provide tools for AI agents operating directly in the browser. This update allows developers to expose existing Model Context Protocol (MCP) tools to web interfaces simply by injecting a single script tag.
Implementation and Proxying Mechanism
According to Vercel documentation, developers can activate this feature by configuring the list of shared tools via the 'experimental_webMcp' object in their server-side codebase. The webpage then loads a script from the corresponding MCP endpoint using the '?webmcp-script' query parameter. This script automatically registers the specified tools within the current webpage environment.
A key architectural highlight is that all tool calls initiated from the browser are automatically proxied back to the MCP server under the logged-in user identity. Thanks to this delegated authorization model, authenticated tools can operate seamlessly without requiring a separate, complex OAuth flow on the client side.
Emerging Web Standard for AI Agents
Currently, WebMCP remains a proposed standard developed under the 'webmachinelearning/webmcp' repository on GitHub, aiming to standardize how AI agents interact with web environments. Vercel encourages developers to upgrade to [email protected] and review the 'WEBMCP.md' technical documentation to begin experimenting with in-browser AI agent integrations.