{"slug": "zephyr-4-5-mcp-server-quic-and-fido2-on-mcus", "title": "Zephyr 4.5: MCP Server, QUIC, and FIDO2 on MCUs", "summary": "Zephyr 4.5, released October 28, ships a built-in Model Context Protocol server that turns every enabled sensor on a networked device into a tool an AI coding agent can call, registering a sensor_read_<node> tool for each Devicetree binding declaring class: sensor. The release also adds QUIC/HTTP3, SSH-2 server and client support, FIDO2 over USB and BLE, RTP (RFC 3550), IPv4 NAT with connection tracking, Infineon TriCore and Qualcomm Hexagon architecture support, 178 new boards, and SPDX 3.0 software bills of materials for EU Cyber Resilience Act compliance. Zephyr reports 21% higher TCP throughput and 12% higher UDP throughput in zperf loopback benchmarks, and documents an agent that read console output from a running sensor driver, diagnosed an interrupt-handler timing bug, applied a fix, and verified it from hardware output without human involvement.", "body_md": "Zephyr 4.5 landed on October 28, and for most of its 10-year history, a new RTOS release meant new board support and driver fixes. This one is different. The standout feature is a built-in MCP server that lets AI coding agents read physical sensors directly from the device — a first in the RTOS world. The release also adds QUIC/HTTP3, SSH remote access, FIDO2 security keys over BLE, GStreamer-style multimedia pipelines, two new architectures, and SPDX 3.0 software bills of materials for EU Cyber Resilience Act compliance.\n\n## MCP Server: AI Agents Can Now Read Your Hardware\n\nThe most interesting addition is not a network protocol or a new chip target. It is the Model Context Protocol server that ships as a first-class Zephyr service. Any Zephyr device with network connectivity can now run as an MCP server, and when it does, every enabled sensor on the board becomes a tool that an AI coding agent can call.\n\nThe implementation leans on Devicetree. When a sensor binding declares `class: sensor`, the MCP server automatically registers a `sensor_read_<node>` tool for it. Claude Code or Codex connects via MCP and gets back every channel that sensor supports — accelerometer axes, gyroscope, magnetometer, barometric pressure, temperature, supply voltages — all with units.\n\nWhat this means in practice: an agent can now close the firmware debug loop on its own. One documented example had an agent connected to a Zephyr device over a serial-port MCP server. It read the console output from a running sensor driver, diagnosed a timing bug in the interrupt handler, applied a fix, and verified the correction directly from hardware output — without a human in the loop. That workflow was not possible before this release.\n\nThis is worth paying attention to beyond the novelty. The boundary between AI-assisted coding and AI-assisted hardware operation has just moved. If you are building IoT products and already use Claude Code or Codex, your toolchain and your firmware can now talk to each other.\n\n## QUIC and HTTP/3: Overdue for IoT\n\nZephyr 4.5 adds a full QUIC transport layer, and the HTTP server can now serve HTTP/3. QUIC ships with TLS 1.3 baked in, establishes connections in fewer round trips, and handles lossy networks significantly better than TCP. That last point matters more for deployed sensors than for most web backends: cellular links, industrial Wi-Fi, and mesh gateways all drop packets routinely.\n\nBeyond QUIC, the IP stack itself is faster. The release reports 21% higher TCP throughput and 12% higher UDP throughput in zperf loopback benchmarks. An RTP stack (RFC 3550) for sending and receiving media over UDP also ships in this release.\n\nAlongside QUIC, 4.5 adds SSH-2 server and client support. SSH into your Zephyr device and you land in the familiar Zephyr shell over an encrypted link — no serial cable required. For teams that field-deploy devices and need secure remote diagnostics, this removes a significant constraint. The release also adds IPv4 NAT with connection tracking and iptables-style forwarding rules, so a Zephyr device can now route traffic for a downstream network independently.\n\n## FIDO2 Over BLE: Your MCU as a Hardware Key\n\nA new FIDO2 subsystem ships with both USB and BLE transport. The BLE transport is the interesting part: a Zephyr device can now act as a hardware security key without being physically connected. An nRF52840-based design becomes a wireless FIDO2 authenticator out of the box, with a sample application included to get you started.\n\n## New Architectures and 178 New Boards\n\nTwo new processor architectures land in 4.5. Infineon TriCore — the architecture inside AURIX automotive microcontrollers — is now supported, opening Zephyr to car ECU development. Qualcomm Hexagon, the DSP found in Snapdragon SoCs, also arrives, relevant for audio and signal-processing workloads on mobile-adjacent hardware. Both come with QEMU support, clocks, interrupts, and serial. One caveat: the TriCore QEMU and Hexagon toolchain are not bundled in the Zephyr SDK yet, so plan for a separate setup step. This release also adds 178 new boards and 29 new shields; more than half the total commits cover drivers and hardware support.\n\n## mpipe: Experimental Multimedia Pipelines\n\nThe mpipe subsystem brings GStreamer-style zero-copy media pipelines to Zephyr firmware. Sources, transforms, and sinks link into pipelines without data copying between stages. The reference demo runs YOLOX-nano object detection with TensorFlow Lite Micro inside Zephyr, detects people, pixelates them, and saves an anonymized image — a working privacy-aware camera on a microcontroller. mpipe is flagged experimental for this release.\n\n## SPDX 3.0 SBOMs for EU CRA Compliance\n\nThe EU Cyber Resilience Act’s reporting obligations went live in September 2026, and Zephyr 4.5 responds with SPDX 3.0 support. Version 3.0 is a ground-up rework of the spec: an SBOM is now a graph of typed elements serialized as JSON-LD, recording the compiler, build flags, prebuilt libraries, and CPE/PURL identifiers for every component. Vulnerability scanners can consume these directly. SPDX 2.3 remains the default; opt in with `west spdx -d build --spdx-version 3.0`.\n\n## Upgrading and Further Reading\n\nIf you are running Zephyr 4.4, the upgrade path is documented in the [official Zephyr 4.5 migration guide](https://docs.zephyrproject.org/latest/releases/migration-guide-4.5.html). The MCP server and SSH additions are the most immediately useful for teams with existing AI-assisted development workflows. QUIC is worth enabling for any device on a cellular or industrial network. SPDX 3.0 matters if EU CRA compliance is on your roadmap. The full feature list and board support tables are in the [official Zephyr 4.5 release announcement](https://www.zephyrproject.org/zephyr-rtos-4-5-now-available-two-new-architectures-multimedia-pipelines-sboms-and-more/) and on the [GitHub v4.5.0 release page](https://github.com/zephyrproject-rtos/zephyr/releases/tag/v4.5.0).", "url": "https://wpnews.pro/news/zephyr-4-5-mcp-server-quic-and-fido2-on-mcus", "canonical_source": "https://byteiota.com/zephyr-4-5-mcp-server-quic-fido2-mcu/", "published_at": "2026-11-19 07:09:56+00:00", "updated_at": "2026-11-19 07:13:51.518541+00:00", "lang": "en", "topics": ["agent-protocols", "ai-agents", "artificial-intelligence", "developer-tools", "ai-tools"], "entities": ["Zephyr", "Model Context Protocol", "Claude Code", "Codex", "Infineon TriCore", "Qualcomm Hexagon", "nRF52840", "EU Cyber Resilience Act"], "also_reported_by": [], "takeaway": "Zephyr 4.5's MCP server auto-registers a sensor_read_<node> tool for every Devicetree sensor binding, letting Claude Code or Codex read accelerometer, gyroscope, magnetometer, pressure, temperature and supply-voltage channels with units and close the firmware debug loop from hardware output.", "recommendation": null, "alternates": {"html": "https://wpnews.pro/news/zephyr-4-5-mcp-server-quic-and-fido2-on-mcus", "markdown": "https://wpnews.pro/news/zephyr-4-5-mcp-server-quic-and-fido2-on-mcus.md", "text": "https://wpnews.pro/news/zephyr-4-5-mcp-server-quic-and-fido2-on-mcus.txt", "jsonld": "https://wpnews.pro/news/zephyr-4-5-mcp-server-quic-and-fido2-on-mcus.jsonld"}}