Your Agent Is an Infinite Canvas — RL Nabors, Dressed for Space
TL;DR
Rachel Lee Neighbors argues that chat interfaces are merely a transitional phase like the CLI was to GUI, demonstrating how HTTP-based MCP servers and interactive MCP apps can turn agents into an 'infinite canvas' for rich web experiences while eliminating inefficient DOM scraping through emerging Web MCP standards.
🌐 MCP Server Architecture & Deployment 2 insights
HTTP transport eliminates CLI configuration friction
Unlike STDIO servers that require users to configure command-line arguments in JSON files, HTTP endpoints only need a URL pasted into agent settings, drastically improving user onboarding for tools like Claude.
Edge functions enable serverless MCP hosting
HTTP MCP servers natively deploy to serverless environments like Vercel and Cloudflare, allowing developers to host agent tools using familiar web infrastructure rather than managing local processes.
🎨 Beyond Chat: Interactive Agent Experiences 2 insights
MCP apps bundle rich media as single HTML islands
By embedding CSS, JavaScript, and base64-encoded assets into isolated iframes, developers can serve full interactive applications like comic readers directly within agent conversations instead of text walls.
Chat interfaces are the new CLI
Neighbors asserts that pure conversational interfaces represent a 'lowest common denominator' transitional phase—much like DOS commands gave way to iPhone touch interfaces—because they force users to possess complete system knowledge without visual guidance.
⚠️ Implementation Constraints & Gotchas 3 insights
Sandbox limitations require embedded design systems
MCP apps lack localStorage, require CSP policies for external fonts/images, and must use call server tool for network requests, making shared component libraries essential for consistent styling.
MCP resources remain unimplemented in major clients
While the MCP spec defines 'resources' for pre-priming agent context with documentation, no current client exposes these in their UI, forcing inefficient workarounds where agents must repeatedly call tools to fetch reference material.
Navigation requires explicit host permissions
External links cannot use standard href or window.open but must request permission via appref current open link, creating a 'mother may I' asynchronous pattern that complicates user flows.
🔮 The Agentic Web & Web MCP 2 insights
Web MCP eliminates DOM scraping inefficiency
Browser-embedded agents currently waste compute power analyzing screenshots or parsing HTML, but upcoming Web MCP standards will allow agents to invoke JavaScript functions directly from visited pages.
Three-client architecture requirements
Modern websites must simultaneously serve humans in browsers, humans using AI agents, and autonomous agents browsing on behalf of users—requiring structured MCP tool endpoints rather than presentation-only HTML.
Bottom Line
Start building HTTP-based MCP servers with interactive MCP apps to make your content accessible to agents, because chat interfaces are transitional and the future belongs to rich, canvas-based agent experiences that eliminate inefficient scraping.
More from AI Engineer
View all
Beyond the Harness: A Journey Towards Adaptative Engineering - Rajiv Chandegra, Annicha Labs
Rajiv Chandegra introduces 'adaptive engineering,' a paradigm shift from fixed AI harnesses (like Cursor or Claude Code) to dynamic, self-organizing systems that emerge during runtime, enabling AI to handle complex, real-world messes beyond deterministic software environments.
What if the harness mattered more than the model? - Aditya Bhargava, Etsy
Aditya Bhargava argues that sophisticated agent harnesses can compensate for weaker open-source models, enabling local AI to match proprietary performance while reducing vendor dependency.
Frontier results, on device - RL Nabors, Arize
Rachel Lee Neighbors introduces a framework for replacing expensive cloud-based frontier models with Small Language Models (SLMs) running on-device, demonstrating how a systematic 'prototype big, deploy small' approach using evaluation tools like Phoenix can cut inference costs to zero while maintaining 90% accuracy and enabling offline functionality.
The Future Is Domain-Specific Agents - Justin Schroeder, StandardAgents
Justin Schroeder argues that the future of AI lies in domain-specific agents—small, specialized agents that compose together rather than general-purpose agents bloated with tools and skills, delivering 80%+ token efficiency and 137x cost savings compared to monolithic approaches.