Adding email sending and Markdown-to-PDF capabilities to an AI agent in just three minutes—the advancements in MCP (Managed Connector Protocol) are indeed insane. But after reading the entire thread, things get interesting.

**Behind the Efficiency Myth**

Santiago demonstrated the possibility of zero-code integration using an MCP server built with xpander.ai. This "no-code victory" narrative is eye-catching. But Levi's吐槽 is more realistic: "MCP is cool, but I found myself writing API scripts directly with the agent." The reason is straightforward: poorly designed MCP servers add too much junk to the context, making them more restrictive than direct connections in certain scenarios.

**Design Determines Success**

Alex's summary hits the nail on the head: "MCP makes integration easier, but lean design is key." It's like putting bulletproof armor on an AI—too heavy and it hampers flexibility; too thin and it loses its protective purpose. Nanbeixidong's observation is even more intriguing: "Navigating the trade-off between MCP's convenience and the fine-grained control of direct API calls is always an interesting journey."

**The Direction of Tool Evolution**

Worth noting is xpander.ai (https://xpander.ai), the full-stack platform Santiago used, which offers a visual workspace, built-in state management, and multi-trigger support. Judging by the product documentation, it aims to solve several core pain points:

1. Pre-configured 2000+ tool connectors, with support for private API customization

2. A version-controlled sandbox to avoid the "crash-on-release" dilemma

3. A built-in Agent Graph system to balance automation and controllability

Robert's comment is spot-on: "MCP is a powerful tool for rapid integration, but maintaining a clean design is crucial." Perhaps this is the norm for technological evolution—every "disruptive innovation" eventually circles back to the most fundamental engineering principles: the right design, for the right scenario.

(Image suggestion: Insert xpander.ai's official Agent Graph system diagram at the end, alongside a comparison with direct API call flows, to create a visual anchor.)