The AI agent interoperability layer is getting a serious upgrade with the launch of Latch, a lightweight script that makes any website instantly usable by AI agents. Built on the emerging W3C WebMCP standard, Latch solves the persistent problem of agents failing to interact with standard HTML buttons and forms. By injecting a single script tag, developers can expose their site’s core functionality as structured tools without touching backend infrastructure.

Zero-Dependency Integration

Latch operates entirely client-side, requiring no server, API keys, or build steps. The script is under 6 KB and loads asynchronously before the closing body tag. It performs read-only discovery of the page to register specific actions an agent needs, such as search_site, add_to_cart, submit_form, and navigate. If the WebMCP standard is not present in the browser, Latch degrades silently, ensuring it never breaks existing functionality.

Supported Browsers and Standards

The tool targets the next generation of AI-native browsers. WebMCP is currently in origin trial for Chrome 149 and Edge 147, with support noted for Atlas and Comet. This timing positions Latch as a forward-looking solution for developers wanting to future-proof their sites for autonomous browsing agents. The script automatically detects search boxes, 'Add to cart' controls, and contact forms, mapping them into callable tools that an agent can execute reliably.

Open Source With Optional Pro Dashboard

The core script is free and open source forever, aligning with the hacker ethos of minimal friction. For teams needing visibility into agent behavior, a Pro tier is available for €19 per month per project. This hosted control plane provides a 90-day history, per-agent breakdowns, and weekly email summaries. Crucially, analytics are opt-in and metadata-only; the script reports tool names and outcomes but never captures form contents, query strings, or PII unless explicitly configured.

Key Takeaways

  • Latch enables WebMCP tool registration via a single