Cloudflare is expanding AI Search into a more managed search layer for agents that need current information from data a developer or organization owns. Instead of manually combining Workers AI, AI Gateway, Vectorize, R2, and Browser Run, users can point AI Search at files or verified websites and let the service handle crawling, ingestion, embedding, and retrieval.
The update adds a discovery option for websites without sitemaps, public search and MCP endpoints across a namespace, custom domains, and an AI Search plugin for sites built on Cloudflare’s EmDash CMS. Cloudflare also previewed pricing in which embedding and reranking are free when customers use selected default Workers AI models.
The company says it uses the system for its own Cloudflare Dev Stack MCP, giving coding agents cited, current documentation from Cloudflare-owned sites rather than stale training data.
The limitation is ownership. Today, website ingestion is built around zones in a customer’s Cloudflare account, with more verification methods planned. That makes it most useful for private or owned knowledge, not general web search.