Why WebMCP matters: agentic web performance and co-browsing
A practitioner argues that site-provided WebMCP tools can make agent-assisted co-browsing faster and more reliable while keeping the user in control of the familiar web experience.
Logged at IST: 2026-10-12 16:29 IST
What it is: Yoav Weiss's practitioner case for WebMCP, written after helping ship Shopify storefront tools.
Gist: Weiss argues that WebMCP lets a site expose explicit JavaScript tools to an agent working alongside the user's browser session. Compared with DOM- or screenshot-driven clicking, those tools can make co-browsing faster, cheaper, and less awkward while leaving the site's developer and the user in control; the user can take over at any point. He calls this “web performance for AI agents.”
He sees WebMCP as a way for a merchant's site to contribute domain-specific help to a user's existing agent, rather than requiring every merchant to build its own assistant. He also argues imperative tools complement declarative/form-based semantics for multi-step tasks and agent-consumed context, while acknowledging unresolved questions around tool registration, permissions, user intent, security, evaluation, and semantic divergence. His cited speed/cost results come from experiments by Nekuda and Shopify, not a universal guarantee.
Newsletter angle: A clear product-level argument for website-first agent integration: expose useful capabilities to the user's agent without replacing the human-facing site or forcing the merchant to own the whole assistant.