Link to Folder with Everything

2026.06.10.officehours.44 - Google Drive

Recording of Video

[https://drive.google.com/file/d/1UMIdhCvptq1q3BQ8zHRUqIFtwnUHgh2j/view?usp=sharing&t=104.619](https://drive.google.com/file/d/1UMIdhCvptq1q3BQ8zHRUqIFtwnUHgh2j/view?usp=sharing&t=104.619)

Summary ( by Gemini )

Discussion covered Model Context Protocol functionality and enterprise security debates while defining standards for agent interaction interfaces.

Defining Model Context Protocol

The group explored the Model Context Protocol as a standardized interface for agents, emphasizing its ability to replace custom integrations with a universal communication layer for AI systems.

Technical Implementation and Capabilities

Participants discussed the technical architecture of Model Context Protocol servers, noting their ability to render interactive user interface elements like dashboards directly within existing chat platforms.

Enterprise Security and Protocols

A debate occurred regarding security boundaries, contrasting Model Context Protocol ease of use against the more rigorous, enterprise-focused security standards of Agent-to-Agent frameworks.

Details

• Intel Craft Organizational Updates: Eric Ramseur initiated a discussion regarding potential MCP (Model Context Protocol) app implementations for Rocket Mortgage, specifically focused on embedding applications and exposing them to 800 million users (00:00:13). Rahul Singh subsequently introduced Intel Craft number 44, emphasizing the community's goal of sharing AI knowledge, inviting presenters to showcase their AI journeys, and encouraging attendees to participate in the Intel Craft WhatsApp community (00:02:09) (00:19:25).
• Engineering Philosophy and Resource Management: Rahul Singh introduced a critique of current AI development practices using a "shovel" metaphor, suggesting that developers often unnecessarily "smelt their own iron" instead of utilizing existing, pre-built components like a "Home Depot" approach to tools (00:07:39). The discussion centered on preventing the reinvention of the wheel by leveraging both open-source and commercial variants of tools that have already been engineered, thereby optimizing time, energy, and resources (00:08:49).
• Model Context Protocol (MCP) Definition: Rahul Singh explained that MCP acts as the "internet for agents," providing a standard that allows AI to communicate with various systems (00:09:47). Before the establishment of MCP, developers relied on unique API integrations using tools like LangChain (now LangGraph); MCP provides a standardized protocol for app-to-agent communication, reducing the need for custom, bespoke integrations (00:11:30).
• Core MCP Architecture and Functionality: Rahul Singh described the architecture of an MCP server, which consolidates and standardizes a menu of capabilities—tools (for actions), resources (for static data), and prompts—for AI agents to utilize (00:13:13). This standardization replaces older function-calling systems, allowing different Large Language Models (LLMs) to interact with the same MCP server without requiring changes to the server logic itself (00:14:26).
• Compatibility and Ecosystem Usage: Rahul Singh highlighted that MCP servers are not limited to chat interfaces; they can be utilized within automation tools like n8n, Dify, and Strands, in addition to interactive chat systems like Claude, ChatGPT, and Cursor. The protocol allows for a diverse range of integrations, ensuring that a single MCP server can serve multiple client types (00:15:39).
• MCP UI and App Demonstration: Rahul Singh provided a practical demonstration using Canva, showing how an MCP server can embed interactive UI elements (such as design grids or lists) directly into the interfaces of ChatGPT and Claude (00:16:44). This implementation allows users to interact with external tools without leaving the chat window, leveraging the iframe-based rendering provided by the protocol (00:18:09).
• Standardization and Terminology: Rahul Singh clarified that while "MCP UI" and "MCP Apps" were originally separate initiatives (an internal library and an OpenAI SDK, respectively), they merged into a single official extension of the MCP protocol in January. While users may hear terms used interchangeably, "MCP Apps" is the preferred standard terminology for building these interactive systems (00:20:39).