Cloudflare's WebMCP Preview: Give Any Site an Agent Interface Without Touching Code
Cloudflare launches a developer preview of WebMCP, letting site owners expose MCP tools to browser-based AI agents via a single toggle, with no origin changes.

Cloudflare has launched a developer preview of WebMCP, a feature that lets any site on its network expose a set of tools for browser-based AI agents. The pitch: flip a switch in the Cloudflare Dashboard, and your site becomes agent-friendly without changing a line of origin code.
WebMCP is an experimental browser standard shipping in Chrome 146, exposed as document.modelContext. It lets sites register tools that agents can call directly, instead of having agents guess their way through human-oriented pages. Cloudflare's implementation injects a small bridge script at the edge using HTMLRewriter, which registers MCP tools with the browser's WebMCP surface.
The bridge script is served from the same origin and loads only if the browser supports WebMCP. It composes tools from packs—groups of related tools—and registers them via .registerTool. Two packs are included in the preview: Content Credentials (C2PA) and Site MCP Server.
The Content Credentials pack scans images for C2PA provenance metadata, returning summaries like claim generator and signing authority. It reads only the first few kilobytes of each image and does not cryptographically verify signatures—every result carries signatureVerified: false to avoid misleading agents.
The Site MCP Server pack proxies tools from your own MCP server, if you have one, letting agents call them through the browser with the visitor's session. All tools run entirely in the browser in this preview, with no round trip to Cloudflare's servers.
Cloudflare also notes that BrowserRun, its remote browser, already supports WebMCP, so you can test your site's tools without writing your own agent. The setup is minimal: enable WebMCP in the dashboard, pick your packs, and the next HTML response includes the bridge tag.
This is a developer preview, and Cloudflare is soliciting feedback via Discord and community forums. The company frames it as a step toward a web that works for both humans and AI agents, without resorting to scraping.
The web was built on the assumption that there is a person on the other end... There is a better way, and it does not involve scraping.