Accessible agent Web? WebMCP || Alex Nahas

What would the web look like if AI agents could use websites alongside people? I sat down with Alex Nahas to discuss WebMCP: a way for websites to expose useful actions to agents without making them navigate every task through screenshots and clicks. We talk about the browser’s existing authentication and security systems, why a website might use WebMCP even when it already has APIs, and why human-facing interfaces still matter. Alex also shares practical advice for developers: expose a small number of powerful tools that match real tasks, rather than turning every button into a separate tool. What would you let an agent do on your website? Subscribe for more conversations about AI, software, and the future of the web. 00:00 Why you shouldn’t expose every action to AI 00:48 Introducing WebMCP and Alex Nahas 01:20 How Alex started working on WebMCP 04:23 The browser and the authentication problem 06:56 What an agent-accessible web looks like 09:32 A web for agents, people, or both? 10:00 Why user interfaces still matter 12:31 “My website already has APIs. Why add WebMCP?” 14:03 What should people control in the interface? 14:22 Which actions should a website expose to agents?