IoT & Embedded MCP Servers — Home Assistant, MQTT, ESP32, ROS, Industrial PLCs, 3D Printers, and More

Part of the IoT & Hardware category.

IoT and embedded MCP servers are bridging the gap between AI agents and the physical world. Instead of writing custom integrations for every device, sensor, and controller, these servers let AI assistants control smart homes, program microcontrollers, monitor industrial equipment, operate robots, and manage 3D printers — all through the Model Context Protocol.

The landscape spans nine areas: smart home platforms (Home Assistant dominates with an official integration plus multiple community servers), smart home devices (Tuya, Philips Hue, Apple HomeKit), IoT platforms (ThingsBoard, AWS IoT SiteWise), MQTT brokers (the backbone of IoT communication), ESP32 and Arduino (AI-driven embedded development), Raspberry Pi (GPIO, I2C, SPI, and camera access), robotics (ROS/ROS2 integration), industrial IoT (OPC UA, Siemens PLCs, SCADA), and 3D printing (multi-platform printer control and STL operations).

The headline findings since our April 2026 refresh: homeassistant-ai/ha-mcp hit 3,300 stars by June with five major releases (v7.4 through v7.7), adding Tool Security Policies, sandboxed custom tool execution via ha_manage_custom_tool, OAuth 2.1, auto-backup before destructive write calls, and a web UI for per-tool enable/disable/pin — by this August audit it had kept shipping through v8.2.0 and grown to roughly 4,400 stars. espressif/esp-claw quintupled to 1,500 stars by June with expanded hardware support (ESP32-P4, C5, and more) and an official M5Stack fork; it’s now roughly 2,000 stars. Espressif launched a hosted documentation MCP server at mcp.espressif.com/docs — real-time AI access to all official ESP chip and SDK documentation. allenporter/mcp-server-home-assistant is now archived — its work was fully merged into Home Assistant Core, which runs at 2026.8.2 as of this audit. voska/hass-mcp (~316 stars) emerged as a strong third option for Home Assistant — it’s a Python project, not Go as we originally reported (see correction below). Ranch-Hand-Robotics/rde-mcp-ros-2 brings 30+ tools embedded in VS Code’s Robot Developer Extension. In 3D printing, DMontgomery40 spun off a Bambu-only fork (now ~110 stars) and codeofaxel/Kiln appeared as a new 900+-tool competitor. midhunxavier/OPCUA-MCP filled the gap left by dormant kukapay/opcua-mcp.

Smart Home Platforms

Home Assistant Official MCP Integration

Home Assistant includes MCP server support as an official core integration (since HA 2025.2, now at 2026.8.2 as of this audit). Any Home Assistant installation can expose its entities, automations, and services to MCP-compatible AI agents without third-party add-ons.

2026.6 update: Previously advanced options are now enabled by default, broadening accessibility. HA now ships two integrations: MCP Server (HA acts as server for external clients) and MCP Client (HA connects out to external MCP servers, including Claude, ChatGPT, Cursor). A read-only home state snapshot resource was added for inspection and debugging. Silver quality classification. Adopted by 3.6% of active HA installations as of this audit — meaningful scale for a relatively new protocol integration. Correction (2026-08-15 audit): the front-matter card description previously stated 2.6% adoption, contradicting this section’s 3.2% — both were stale; the live figure on Home Assistant’s own integration page is 3.6%, now reconciled throughout this page.

homeassistant-ai/ha-mcp (Community Leader — ~4,400 Stars)

The fastest-growing and most feature-rich Home Assistant MCP server — and the one with the most aggressive release cadence in the entire IoT category. Grew from 2,500 to 3,300 stars (+32%) between our April and June refreshes, with five major releases in that window:

• v7.4.0 (Apr 29): Unified helper deletion across 27 types, energy preferences tool, addon API renamed to ha_manage_addon
• v7.5.0 (May 13): Sandboxed code execution via ha_manage_custom_tool — AI agents can create and run custom tools within a safe sandbox; scene config tools; event bus publishing; web UI for per-tool enable/disable/pin; OAuth 2.1 mode (beta)
• v7.6.0 (May 27): Tool Security Policies — per-tool approval gating so sensitive operations require explicit user sign-off; configurable HTTP bind host; auto-backup before destructive write calls; Assist pipeline management
• v7.7.0 (Jun 10): User-configurable filesystem directories; consolidated search; nested beta toggle panel; improved automation/script YAML handling

Since then (this August audit): the pace has not slowed — releases have continued through v8.2.0, including a security-policy rework, single-source translations with new locales (Italian, Spanish, Swedish, Polish), and a secrets-redaction toggle. Stars have grown to roughly 4,400 and the tool count to 95+.

The Agent Skills system (introduced in April’s v7.3.0) remains a differentiator — domain-specific guides served as MCP resources that teach AI agents how to write automations, select helpers, and safely refactor HA configurations. The Webhook Proxy add-on routes MCP traffic through existing reverse proxies with no separate tunnel needed.

tevonsb/homeassistant-mcp

The original well-established community server, now in slow maintenance mode. Star count essentially flat (554 → 575). No major new releases since April 2026. Still useful as a stable, minimal alternative — but ha-mcp has superseded it for feature depth. Correction (2026-08-15 audit): license was previously listed as MIT; the repository’s actual license is Apache-2.0.

voska/hass-mcp (NEW — ~316 Stars)

A strong new entrant. Focuses on entity management, device history, statistics, and guided conversations, and requires Python 3.13+ with uv. Correction (2026-08-15 audit): this entry previously described the server as Go-based with HTTP transport, calling it “the only major HA MCP server in Go.” That was wrong — the repository’s README and packaging (pyproject.toml, uvx hass-mcp) show it is a Python project, MIT-licensed. Apologies for the error; the “Go” framing has been removed.

allenporter/mcp-server-home-assistant (Archived)

Archived on March 2, 2025 — read-only. This project served its purpose: the custom component was upstreamed into Home Assistant Core via PR #134122. Superseded by the built-in HA integration. No longer relevant for new deployments.

tdeckers/openhab-mcp

MCP server for openHAB — the other major open-source smart home platform. Interacts via its REST API. Lower adoption than Home Assistant servers but fills an important gap for openHAB users.

Smart Home Devices

tuya/tuya-mcp-sdk (Official)

Official SDK from Tuya — one of the world’s largest IoT cloud platforms. Provides Python, Go, and C# SDKs for integrating Tuya Cloud capabilities with AI agents via MCP. Tuya’s AI Agent Development Platform fully supports MCP server integration. Correction (2026-08-15 audit): previously described as “essentially flat since April” at ~16 stars — it has since grown to 65 stars, so that characterization no longer holds.

A community-built Tuya Smart Home MCP Server (4,500+ downloads) adds local control via tinytuya for lower latency and better privacy.

rmrfslashbin/hue-mcp

Modern MCP server for Philips Hue lighting systems. Supports multi-bridge control, real-time sync, and comprehensive tools for lights, rooms, scenes, and bridges. The Go implementation is notable — most IoT MCP servers are Python.

ykhli/mcp-light-control

Philips Hue light control for Cursor and Claude Desktop — including sending messages through lights using Morse code. A creative demonstration of AI-hardware interaction.

thomasvincent/home-mcp (Repository Gone)

Correction (2026-08-15 audit): github.com/thomasvincent/home-mcp now 404s. We could not find a working repository at that URL or at a plausibly-renamed owner we checked (somethingwithproof/home-mcp, also 404). We’re removing the entry rather than link a dead repo. Readers looking for Apple HomeKit MCP control should search current options directly — several newer HomeKit-focused MCP servers (e.g. from other authors) have appeared since this entry was originally written, but we have not vetted them and won’t name one without verifying it first.

IoT Platforms

thingsboard/thingsboard-mcp (Official — Most Tools)

Official MCP server from ThingsBoard — the most popular open-source IoT platform. Stars grew significantly (from ~20-50 range to 98). With 120+ tools, this has the highest tool count of any server in the IoT/embedded category: device provisioning, entity management, telemetry analytics, dashboard creation, and operational automation. Correction (2026-08-15 audit): language was previously listed as Python; the repository is built in Java (the ThingsBoard platform’s primary language).

v2.1.0 (February 2026): Added API key authentication, tool groups for context window optimization, OTA package management, device creation tools, and improved cross-LLM tool descriptions. No new commits since February 2026 — the project is well-featured but currently quiet.

ThingsPanel/thingspanel-mcp (Official)

Official MCP server from ThingsPanel — another open-source IoT platform. Supports MQTT and Modbus devices with integration for Claude and GPT models.

AWS IoT SiteWise MCP (Official)

Official AWS MCP server for IoT SiteWise — AWS’s industrial IoT service. 20+ tools for asset management, data ingestion, monitoring, analytics, time-series aggregations, and bulk export via S3 with KMS encryption. Part of the awslabs/mcp monorepo (~9,600 stars as of this audit, up from 9,254 in June and 8,500 in April).

Duke-CEI-Center/IoT-MCP-Servers

Academic IoT MCP server from Duke University‘s Center for Emerging Interface Center. Reads sensor data and dispatches data collection tasks for AI assistants. Note the non-commercial license (CC BY-NC-SA 4.0) — unlike most servers on this page, it is not free to use commercially.

MQTT

MQTT is the dominant messaging protocol in IoT — lightweight, pub/sub, designed for constrained networks. Multiple MCP servers bridge AI agents to MQTT brokers.

ezhuk/mqtt-mcp

Lightweight MCP server connecting LLM agents to MQTT devices. Designed for building automation, industrial control, and smart home systems. Uses FastMCP 2.0.

Benniu/emqx-mcp-server

MCP server for EMQX — the most popular open-source MQTT broker. Supports both EMQX Cloud and self-hosted clusters.

CorefluxCommunity/Coreflux-MQTT-MCP-Server (Official)

Enterprise-grade MCP server from Coreflux for their MQTT broker. Full TLS support, API access, AI code generation, and health monitoring. 15+ tools — the most feature-rich MQTT-specific MCP server.

tspspi/mcpMQTT

Generic MQTT interface with fine-grained topic permissions using wildcard matching. Security-conscious design for multi-tenant MQTT environments.

mqtt-ai/mcp-over-mqtt

An MCP-over-MQTT transport specification with built-in service registry, discovery, and load balancing. Enables MCP in IoT environments where HTTP isn’t practical.

ESP32, Arduino, and Embedded Frameworks

The embedded MCP space continues to accelerate. Espressif — the company behind ESP32 — has made the most visible vendor commitment of any hardware manufacturer to MCP, and that commitment deepened significantly since our April refresh.

espressif/esp-claw (Official — AI Agent Framework — ~2,000 Stars)

The biggest star-growth story in the IoT category since our April refresh: ESP-Claw quintupled from 304 to 1,500 stars by June, and has since grown further to roughly 2,000 as of this August audit. Espressif’s official “Chat Coding” AI agent framework for ESP32 devices acts as both MCP server and client — exposing hardware capabilities to external agents while calling external services.

Key features: on-device memory (structured long-term memory lives on-chip, preferences and routines auto-extracted from conversations), Chat Coding (define device behavior through natural conversation, with LLM handling dynamic decisions and local Lua scripts executing deterministically even offline), and support for OpenAI, Anthropic, DeepSeek, Qwen, and custom LLM endpoints.

Hardware support has expanded: originally ESP32-S3 only, now also supports ESP32-P4, ESP32-C5, ESP32-S31, and a variety of dev boards including M5Stack CoreS3. M5Stack has officially forked esp-claw for their device line — a significant sign of ecosystem momentum. Last updated June 2026.

ESP-IDF 6.0 Built-in MCP Server (Official)

ESP-IDF 6.0 (released March 2026) ships with a built-in MCP server for development workflows. Tools cover building, flashing, setting the target, and cleaning, plus resources for querying project configuration, build status, and connected devices. Launched via eim run using the ESP-IDF Installation Manager.

The official embedded development framework for the world’s most popular IoT microcontroller now treats MCP as a first-class feature.

Espressif Documentation MCP Server (Official Hosted — NEW)

Espressif launched a hosted documentation MCP server at https://mcp.espressif.com/docs (announced April 8, 2026). Authentication via GitHub or WeChat. Connects AI agents to all official Espressif documentation in real time — ESP-IDF, ESP chips, SDKs. Read-only; does not execute code or modify files.

Combined with the ESP-IDF 6.0 built-in tools server, this gives AI agents a complete ESP32 development workflow: look up docs via the hosted server, build and flash via the local tools server.

espressif/esp-rainmaker-mcp (Official — IoT Device Control)

Official MCP wrapper for ESP RainMaker — Espressif’s cloud IoT platform. Bridges MCP clients with the RainMaker CLI for natural language control of IoT devices via Claude, Cursor, Gemini CLI, and Windsurf. Correction (2026-08-15 audit): star count was previously described as flat (~9 → ~7); it has since grown to 17.

78/xiaozhi-esp32 (MCP-Based Chatbot — ~28.9K Stars)

The most starred project in the IoT MCP space. Xiaozhi is an open-source MCP-based chatbot running on ESP32 hardware. The v1 branch was retired in February 2026. Recent updates (v2.2.6, April 2026) included new hardware board support (Waveshare ESP32-S3-Touch-LCD, M5Stack Cardputer Adv), wake word fixes, LCD layout improvements, and camera/audio optimization; more recently the project migrated to ESP-IDF 6.0 (v2.4.0, July 2026) and shipped v2.4.2 (Aug 2026), the current latest release. Now supports 70+ boards with cloud-side MCP for smart home control, PC desktop operation, and knowledge search. Correction (2026-08-15 audit): star count and firmware version were stale (~24,000 stars / v2.2.6, both from the June refresh); updated to current figures, and the license field (previously blank) is confirmed MIT.

While technically an MCP client rather than server, its ~28.9K stars and active community make it the most visible demonstration of MCP in the embedded world.

horw/esp-mcp

The original community ESP-IDF MCP server. Centralizes ESP-IDF commands for AI-driven interaction — build, flash, and automatic issue fixing. Star count flat since April. Still useful as a more configurable alternative to the official built-in server.

rzeldent/esp32-cam-ai (NEW)

New MCP server for ESP32-CAM modules: image capture, LED/flash control, and system diagnostics. Connects AI agents directly to ESP32 cameras for vision tasks without a separate Pi or compute board.

jurgen178/esp32-mcp (NEW)

Notable for running MCP natively on Arduino Nano ESP32 entirely in C++ — without an official C++ MCP SDK. A proof-of-concept that MCP can run server-side on very constrained hardware using only the Arduino framework.

MicroPython MCP Server

An MCP bridge server allowing direct code execution from an LLM on any board running MicroPython (ESP32, RP2040, etc.) via USB Serial or WebREPL. Compatible with Claude Desktop, VS Code Codex, Copilot, and Antigravity.

solnera/esp32-mcpserver

A lightweight MCP server framework for ESP32, available on PlatformIO. Seamlessly connects embedded devices to LLMs via WebSocket/JSON-RPC. Developers build custom tools for their specific hardware.

Serial Communication Servers

Three serial communication MCP servers for talking to Arduino, ESP32, and other hardware over USB serial. mcp2serial (49 stars) leads with Raspberry Pi Pico support. serial-mcp-server provides the most comprehensive serial capabilities with professional debugging features and JSON macro automation. serial-mcp is a minimal implementation built specifically for Claude Code. Correction (2026-08-15 audit): serial-mcp-server was previously listed as Python and 10-20 stars; the repository is actually written in Rust and now has 85 stars — both the language and the star count were wrong/stale.

Raspberry Pi

grammy-jiang/RaspberryPiOS-MCP

The most comprehensive Raspberry Pi MCP server. Modular design with 15+ tools covering:

• System metrics — CPU, memory, disk, temperature monitoring
• Service management — start, stop, restart systemd services
• GPIO — digital I/O via gpiozero
• I2C — sensor communication via smbus2
• SPI and serial — hardware bus access
• Camera — Pi camera control and capture

UnitApi/mcp

Secure hardware control through MCP with real-time GPIO streaming. Supports LED control, button input, and sensor operations on Raspberry Pi.

Robotics (ROS)

robotmcp/ros-mcp-server (Most Starred IoT MCP Server)

The most starred IoT MCP server. Provides bidirectional AI-robot communication across both ROS1 and ROS2: natural language → robot commands (topic publications, service calls, action goals), and robot state → AI understanding (topics, transforms, sensor data). Grew from 1,200 (April) to 1,300 (June) to roughly 1,400 as of this August audit.

wise-vision/ros2_mcp

An advanced ROS 2 MCP server with 1-minute setup, auto type discovery, image streaming (Image and CompressedImage), and full action support. Star count grew from ~60 (June) to 86 (this August audit); license, previously blank, is confirmed MPL-2.0. wise-vision had also published this as mcp_server_ros_2 on the Docker MCP Catalog — Correction (2026-08-15 audit): that repository now 404s and could not be verified; treat the Docker MCP Catalog listing as unconfirmed until re-checked.

Ranch-Hand-Robotics/rde-mcp-ros-2 (NEW)

A new ROS 2 MCP server embedded in the Robot Developer Extension for VS Code. 30+ tools for full system introspection — topics, services, actions, nodes, parameters, and logs. Designed for IDE-native robot development with AI assistance. The VS Code integration is notable: developers get AI-assisted ROS 2 development without leaving their editor.

lpigeon/ros-mcp-server

ROS/ROS2 MCP server focused on transforming natural language into robot commands. Actively maintained.

Additional ROS Servers

Industrial IoT — OPC UA, PLCs, and SCADA

midhunxavier/OPCUA-MCP (NEW — Most Capable OPC UA Server)

The new leader for OPC UA MCP integration. Dual Python and Node.js implementations, supporting browse nodes, read/write variables, call methods, historical data retrieval, and batch operations. Has effectively superseded kukapay/opcua-mcp as the go-to community OPC UA server. Star count grew from 13 (June) to 16 (this August audit); license field, previously blank, is confirmed MIT.

kukapay/opcua-mcp

MCP server connecting to OPC UA-enabled industrial systems — reads and writes OPC UA node values. Dormant since October 2025 — no updates in 2026. Star count crept up to 29 but the project appears stalled. midhunxavier/OPCUA-MCP is the more actively maintained alternative.

cadugrillo/s7-mcp-bridge

MCP server connecting AI agents to Siemens S7-1500 and S7-1200 PLCs using the S7 protocol. Dormant since January 2026 — no updates for several months. The Go implementation and S7 protocol support remain unique, but the project is stalled. License field, previously blank, is confirmed MIT.

poly-mcp/IoT-Edge-MCP-Server (Repository/Org Gone)

Correction (2026-08-15 audit): github.com/poly-mcp/IoT-Edge-MCP-Server now 404s, and so does the poly-mcp organization page itself and its other listed repo (poly-mcp/PolyMCP) — the whole namespace appears to have been deleted or renamed off GitHub since our June refresh. We searched for a documented successor and found none we could verify (a similarly-named AndreiT0/IoT-Edge-MCP-Server also 404s and its provenance is unconfirmed), so we’re removing this entry rather than link a dead or unverifiable repo. It previously described an enterprise-grade Industrial IoT server unifying MQTT sensors and Modbus devices; readers looking for that kind of functionality should search GitHub directly for current MQTT/Modbus industrial MCP bridges rather than rely on this now-dead link.

AddisonTech/Hermes (NEW)

Very new (updated June 11, 2026). Rust-based bridge connecting AI agents to plant-floor industrial data via OPC-UA. Noteworthy as the only Rust-based OPC UA MCP server — early stage but potentially interesting for latency-sensitive industrial use cases.

3D Printing

DMontgomery40/mcp-3D-printer-server

The most comprehensive multi-platform 3D printing MCP server. Grew from 168 stars (April) to 195 (June) to 225 as of this August audit. Connects to 8 printer platforms: OrcaSlicer, Bambu Lab, OctoPrint, Klipper, Duet, Repetier, Prusa, and Creality.

Recent updates (v1.2.4, v1.2.6 — May 2026): Added FULU OrcaSlicer/Bambu support, fixed HTTP compatibility entrypoints, and fixed a schema bug in upload_gcode inputSchema for Anthropic tool compatibility. Available on Docker Hub and npm.

The maintainer also spun off DMontgomery40/bambu-printer-mcp (see below) as a leaner Bambu-only alternative.

DMontgomery40/bambu-printer-mcp (Bambu-Only Fork — ~110 Stars)

A spin-off from the same author. Strips all non-Bambu platforms for a leaner, faster install focused entirely on Bambu Lab printers. Created May 2026; grew from 57 stars in June to roughly 110 as of this August audit — nearly doubling. Includes STL manipulation, BambuStudio slicing, AMS management, camera snapshots, and (per its latest release) AMS dryer control. Good choice for Bambu-only shops that don’t need multi-platform overhead.

codeofaxel/Kiln (NEW — Most Ambitious New Entrant)

The most ambitious new 3D printing MCP server. Its own docs put the count at 901 tools plus prompts/resources for 908 total MCP capabilities — more than any other server in the 3D printing category, and higher than the 810+ we originally reported. End-to-end text-to-print workflows with multi-printer support (OctoPrint, Klipper, Bambu, Creality, Prusa, AnyCubic, Elegoo, USB/Marlin), failure recovery with layer-level resumption, and marketplace integration (MyMiniFactory, Cults3D). Includes outsourcing via Craftcloud. Freemium: free tier (unlimited local printing, one printer), Pro at $49/month (branching, cloud sync, mid-print modification), Business at $199/month (commercial use, multi-printer fleet management), and a separate Enterprise tier. Backed by Hadron Labs Inc. Star count grew from 21 (June) to 46 (this August audit). Correction (2026-08-15 audit): license and language were previously left blank; the repo is AGPL-3.0-licensed (with commercial licensing available separately) and primarily Python. Worth watching — the scope is far beyond any other 3D printing MCP server.

OctoEverywhere/mcp (Official)

Official MCP server from OctoEverywhere — the most popular cloud management platform for 3D printers. Supports OctoPrint, Klipper, Bambu Lab, Elegoo, Prusa, and Creality. Provides live printer state, webcam snapshots, and printer control. Code-frozen since July 2025 — the server remains in production but no new features have shipped in nearly a year. The cloud backend continues to improve, but the MCP server layer is stable/static.

schwarztim/bambu-mcp

Complete MCP server for Bambu Lab 3D printers. 25 tools covering print control, status monitoring, camera access, AMS, temperature monitoring, and LED control. Local MQTT communication with X.509 certificate authentication. Quiet since April 2026.

Disane87/spoolman-mcp (NEW)

A new adjacent tool: filament inventory management via the Spoolman API. Not a printer controller — manages vendors, filament types, and spool inventory with full CRUD, JSON/CSV export, and database backup. Useful alongside any of the printer control servers for tracking material consumption.

Other Notable Servers

Drone Control

MCP server for controlling DJI Tello drones — takeoff, landing, movement commands, and video streaming.

Bluetooth

Bluetooth device scanning and interaction. Supports both BLE and Classic Bluetooth on Windows, macOS, and Linux.

Zigbee2MQTT

MCP server for Zigbee2MQTT with intelligent device discovery — auto-analyzes all connected Zigbee devices and exposes their capabilities.

Node-RED Integration

Two approaches to Node-RED MCP integration: node-red-contrib-mcp adds MCP nodes to Node-RED’s visual flow editor; node-red-mcp-server lets AI agents manage Node-RED flows, nodes, and settings. Correction (2026-08-15 audit): node-red-mcp-server’s star count was stale (<10 → actually 39), and the repository has since been archived (May 21, 2026) — it’s now read-only and no longer actively maintained.

Sensor Monitoring

MCP server for the Aranet4 CO2 sensor. Scans for devices, fetches data, stores in local SQLite, and provides historical querying with visualization.

Curated Resources

beriberikix/awesome-mcp-hardware — A curated awesome-list specifically for MCP servers interacting with hardware and the physical world. Good reference for discovering new IoT/embedded MCP servers as they emerge.

What’s missing

The gap list is largely unchanged from our April refresh:

• No official Arduino IDE integration — the most popular embedded development platform has no MCP server for project management, board configuration, or compilation (Espressif’s ESP-IDF 6.0 built-in server partially addresses this for ESP chips only)
• No major Zigbee/Z-Wave hub vendors — Samsung SmartThings, Hubitat, and Aeotec have no official MCP servers
• No PLC vendor servers — Siemens, Allen-Bradley (Rockwell), Mitsubishi, and ABB have no official MCP offerings (only dormant community bridges)
• No Matter/Thread protocol servers — the new smart home standard has no dedicated MCP support (accessible indirectly through Home Assistant)
• No LoRaWAN servers — the dominant LPWAN technology for IoT remains absent from the MCP ecosystem
• No edge AI servers — no MCP integration for TensorFlow Lite, ONNX Runtime, or other edge inference frameworks (ESP-Claw’s on-device inference is closest)
• Limited sensor aggregation — individual sensor servers exist but no unified multi-sensor hub

The bottom line

Rating: 4.0/5 — The IoT/embedded MCP ecosystem is deepening, not just widening. The home automation subcategory stands out: homeassistant-ai/ha-mcp hit 3,300 stars with five releases in under seven weeks by June, added enterprise-grade features (Tool Security Policies, sandboxed execution, OAuth 2.1) that put it ahead of most non-IoT MCP servers in sophistication, and has kept shipping since — roughly 4,400 stars and v8.2.0 as of this August audit. The allenporter project completing its mission — merging into Home Assistant Core and gracefully archiving — shows what healthy open-source MCP development looks like.

Espressif’s story is the other big headline: ESP-Claw quintupled in stars by June (from 304 to 1,500) and has continued growing since (roughly 2,000 as of this audit), ESP-IDF 6.0 ships MCP as a built-in feature, and a hosted documentation server at mcp.espressif.com/docs completes a three-part official AI development stack for ESP32. M5Stack forking ESP-Claw signals that the embedded AI agent paradigm is moving downstream from experiments to product.

The remaining gaps — vendor adoption from Siemens, Arduino, and SmartThings; emerging standards like Matter and LoRaWAN — are consistent. The category will hit 5.0/5 when industrial PLC vendors and smart home hub makers ship official servers. Until then, 4.0/5 reflects a well-developed ecosystem with meaningful but still-present holes.

This review was refreshed on 2026-06-12 using Claude Sonnet 4.6 (Anthropic); citations and figures re-verified 2026-08-15 using Claude Sonnet 5 (Anthropic). Originally published 2026-03-15.