MCP-Edge
Integration with Arduino devices and Arduino IoT Cloud, enabling management of device properties and firmware-in-the-loop testing with Arduino, ESP32, and RP2040 boards.
Exposes Edge Impulse ML inference as MCP tools, allowing AI agents to run models on edge devices.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
MCP-Edge
Extend the Model Context Protocol (MCP) to edge and IoT devices.
MCP-Edge lets cloud LLM agents discover and invoke physical hardware through the same tool interface they already use for software APIs. It provides:
a gateway that bridges cloud-native MCP transports (SSE / HTTP) to constrained device channels (UART, BLE, local Wi-Fi), presenting every downstream device as a standard MCP tool provider;
MCP-Lite, a lightweight MCP server for devices with as little as ~512 KB of RAM;
a four-tier device taxonomy (constrained MCUs, smart IoT nodes, BLE-only wearables, Linux-class edge computers) that maps an MCP strategy to each tier;
protocol adaptations for constrained links: CBOR encoding, schema caching, connection pooling, and offline request buffering.
Status: alpha, under active development. The public API is unstable and will change. This is the reference implementation of the MCP-Edge framework (IEEE Cloud Summit 2026): the gateway, MCP-Lite server, four-tier device taxonomy, and serial/UART, BLE, and Wi-Fi (TCP) transports, all covered by a hermetic test suite, with ESP32 MicroPython example firmware and a Wokwi boot test in CI, and the gateway install + boot verified on aarch64 Linux and Raspberry Pi OS (QEMU) in CI — see platform support.
Installation
Requires Python 3.10+.
pip install mcp-edgeFor development, install from source:
git clone https://github.com/jemsbhai/mcp-edge
cd mcp-edge
pip install -e ".[dev]"Related MCP server: Jeltz
Quickstart
Run a demo gateway — an MCP server exposing two simulated devices over stdio:
mcp-edge run --demoThis serves a simulated sensor (read_temp, read_humidity) and a simulated ring
(heart_rate), each tool namespaced by device (sensor-01/read_temp, ...). To drive it
from an MCP client such as the MCP Inspector or Claude Desktop, configure a stdio server
with command mcp-edge and arguments ["run", "--demo"]. The process logs to stderr and
waits for a client on stdin.
Use the gateway as a library:
import asyncio
from mcp_edge.client import MCPLiteClient
from mcp_edge.devices import SimulatedDevice
from mcp_edge.gateway import Gateway
from mcp_edge.registry import DeviceRegistry
from mcp_edge.tiers import Tier
from mcp_edge.transports import LoopbackTransport
async def main() -> None:
device = SimulatedDevice("sensor-01")
device.add_tool("read_temp", lambda args: {"celsius": 21.5})
transport = LoopbackTransport(device.handle)
await transport.open()
registry = DeviceRegistry()
registry.register("sensor-01", MCPLiteClient(transport), Tier.SMART_NODE)
gateway = Gateway(registry)
print([tool["name"] for tool in await gateway.list_tools()]) # ['sensor-01/read_temp']
print(await gateway.call_tool("sensor-01/read_temp", {})) # {'celsius': 21.5}
asyncio.run(main())Connect your own device over serial (experimental) — install the serial extra
(pip install "mcp-edge[serial]") and point SerialTransport at the port your device is
on. It exposes the same Transport interface, so it drops into the example above in place
of LoopbackTransport:
from mcp_edge.transports import SerialTransport
# "COM3" on Windows, "/dev/ttyUSB0" on Linux/macOS
transport = SerialTransport("COM3", baudrate=115200)Frames use a length-prefixed wire format — a 2-byte big-endian length followed by the CBOR
payload — documented in transports/serial.py. Device firmware must frame its replies the
same way; the exported encode_frame / decode_frame helpers make that straightforward.
This path is not yet validated on physical hardware. Worked ESP32 MicroPython firmware
for the serial, BLE, and Wi-Fi transports lives in
examples/firmware/esp32_micropython/.
Roadmap
v0.1 — gateway core, in-process (loopback) transport, protocol adaptations (CBOR, schema caching, connection pooling, offline buffering), device simulator, health monitor, CLI, hermetic CI
v0.2 — real transports and Wokwi firmware-in-the-loop tests (Arduino / ESP32 / RP2040). The serial/UART (
pyserial), BLE (bleak), and Wi-Fi (TCP, with mDNS discovery viazeroconf) transports have landed (software-tested), each with ESP32 MicroPython example firmwarev0.2+ — Edge Impulse (inference as an MCP tool) and Arduino IoT Cloud (properties as MCP) integration examples; Renode / QEMU backends
Development
pip install -e ".[dev]"
pytest -q
ruff check .License
MIT — see LICENSE.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
One MCP endpoint for Claude, GPT & Gemini: 100+ tools + no-code connectors + agent workers.
Connect any AI agent to 1,000+ apps and 27,000+ actions through one remote MCP server (OAuth).
The OpenRouter for tools. One MCP connection gives any AI agent 254 hosted tools, pay per call.
Zero-setup MCP gateway securely connecting AI to your tools with authentication and workflows
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceEnables LLM agents to securely monitor and control MQTT devices for building automation, industrial control, and smart home systems through a standardized MCP interface.25MIT
- AlicenseNot gradedqualityDmaintenanceConnects AI assistants to physical devices via MCP, enabling LLMs to reason about sensor data, perform cross-device correlation, and interpret anomalies.4Apache 2.0
- AlicenseAqualityDmaintenanceEnables AI agents to discover, monitor, and control industrial hardware like ESP32 machine controllers and inspection cameras through a standardized MCP interface.5MIT
- AlicenseNot gradedqualityBmaintenanceEnables AI agents and humans to control various hardware boards (ESP32, Arduino, Raspberry Pi, etc.) through a unified API via MCP, with real-time dashboard support.MIT