Agentic AI at the Edge: Bringing Model Context Protocol (MCP) to Microcontrollers with Zephyr

In this episode, NXP Systems Enablement Engineer David Piskula talks about Agentic AI at the edge and the Model Context Protocol (MCP), and how NXP is bringing MCP down to the microcontroller level using Zephyr. The discussion covers why edge computing matters (security, confidentiality, low latency), how MCP standardizes JSON message exchanges between clients and servers, and how self-describing "tools" enable plug-and-play device integration. Critical in scenarios like smart homes and factory automation. David shares the challenge of implementing MCP on resource-constrained MCUs. He highlights the benefits of Zephyr providing a hardware agnostic path for evaluation. The NXP solution is available now to evaluate with an LED-control demo using NXP's eIQ Agentic AI framework.

Other episode resources:

• LED-Control Demo project (GitHub repo)
• Zephyr Upstream mcp_server contributed by NXP
• www.nxp.com/zephyr

00:00 Welcome to EdgeVerse

01:07 Meet David Piskula

01:47 What Systems Enablement Does

03:09 Why Edge AI Matters

04:15 MCP Explained

04:56 How MCP Works

05:45 Smart Home Example

06:45 Factory Automation Use Case

07:26 MCP on Microcontrollers

08:32 Biggest Engineering Challenges

09:31 Plug and Play LED Demo

11:11 How to Get Started

12:22 Wrap Up and Next Steps