Build agent-ready sites with WebMCP

YouTube AI Channels Tools

Summary

OpenAI introduces WebMCP support for Codex, a protocol enabling websites to expose tools directly to AI agents for autonomous interaction, as demonstrated through a collaborative 3D modeling studio.

No content available
Original Article
View Cached Full Text

Cached at: 08/26/26, 05:36 AM

# Build Agent-Ready Sites with WebMCP: Enabling Codex to Interact with Websites **TL;DR:** OpenAI introduces WebMCP support for Codex, a protocol that allows websites to expose tools directly to the AI agent, enabling it to autonomously understand and use site capabilities for complex tasks, as demonstrated through a collaborative 3D modeling studio. ## Introduction: Moving Beyond Clicking to Integrated Tools Codex has been capable of interacting with the web for some time, and watching it navigate sites by clicking around is often impressive. However, this method has a significant drawback: it can be quite slow. To solve this, OpenAI is launching WebMCP support for Codex. WebMCP is a protocol that allows you to integrate tools directly into your website. This enables Codex to autonomously discover and understand the capabilities your site offers, allowing it to achieve excellent results within your specific environment. ## Demonstrating the Power: The "Codex Modeling Studio" To showcase the potential of WebMCP, a small demo called "Codex Modeling Studio" was built. This demo allows Codex to open a website and visually iterate on a 3D model—a model that could potentially be 3D printed later. Codex can set shapes and take screenshots from different angles without disrupting the user's view. In this demo, Codex and the user can collaborate, with the user able to provide notes as Codex works. This small demo illustrates a powerful concept: you can embed advanced functionalities directly into a website. Websites are no longer limited to simple e-commerce pages; they can now include full 3D rendering and complex features. The integration of WebMCP significantly enhances how you can present and distribute content to users. ## Implementing WebMCP: A New Paradigm for Tool Development Adding WebMCP to your website is designed to be as simple as letting Codex operate on it. Codex can autonomously understand the specification and write the tools for you. The process involves treating Codex as your "client"—the primary user of the tools, not a human. Therefore, you must ensure the tools are discoverable by Codex and that its objectives are clear. It is recommended to spend time testing and iterating with Codex. You can refine the toolset, ensure descriptions are clear, and help Codex complete tasks as efficiently as possible. Codex can provide feedback when requested; for instance, the demo includes a feedback tool that allows Codex to flag errors and alert the user when fixes are needed. ## A New Experience: Models and Humans Collaborating on the Same Interface Building tools for a model is fundamentally different from building tools for humans. However, we have now entered a new stage where models and humans can collaborate on the same interface through different channels. This represents a completely new experience. The ability to have Codex build its own tools and to collaboratively iterate on visual content—especially when you see something different from what Codex sees—is an entirely new frontier. The potential of agents independently interacting with websites is just beginning to be explored, and the possibilities for future creations are vast. Source: [YouTube Video](https://www.youtube.com/watch?v=Is2NHa7awWY)

Similar Articles

WebMCP: Teaching Your Website to Talk to AI Agents

Hacker News Top

WebMCP is a proposed web standard developed by Google and Microsoft that allows websites to declare structured tools for AI agents to call directly, replacing fragile screen-scraping with stable interfaces.

@danshipper: Codex native apps are here!

X AI KOLs Timeline

Codex native apps have been released, and OpenAI has published a blog post about building with WebMCP for collaborative UIs between agents and humans.