Descriptions:
Rachel Lee Neighbors — Principal Developer Experience Engineer at Arise and former contributor to Mozilla Firefox DevTools, the W3C Web Animations API, and Microsoft Edge — delivers a conference talk on making web content navigable by AI agents using browser primitives that already exist in the current spec. Drawing from her own real-world project of reviving a 2000s-era webcomics archive (rachelagreat.com), she walks through building MCP (Model Context Protocol) servers that expose structured content to three distinct client types: humans in browsers, humans using agents, and autonomous agents browsing on behalf of users.
The talk covers MCP tool design and transport hosting, demonstrating tools for listing comics, story arcs, characters, and transcript-powered search. A standout section shows how MCP tools can return not just plain JSON or markdown but fully interactive UI components — embedding a comic reader directly inside an agent interface rather than forcing users to consume walls of text in chat.
Neighbors also delivers a sharp critique of the current MCP ecosystem: the spec’s “resources” feature, intended to let clients pre-load context such as documentation without requiring individual tool calls, remains too loosely defined and poorly implemented across agent harnesses. She calls on client developers to provide even bare-bones resource support. The talk is a practical blueprint for any developer looking to make existing web properties agent-accessible today, with code available in the linked repository.
📺 Source: AI Engineer · Published May 23, 2026
🏷️ Format: Tutorial Demo
