cd /news/ai-tools/show-hn-i-let-an-ai-agent-brick-my-2… · home topics ai-tools article
[ARTICLE · art-103468] src=github.com ↗ pub= topic=ai-tools verified=true sentiment=· neutral

Show HN: I let an AI agent brick my $2.51 ESP32 on purpose

NanoForgeFlow released nff, an open-source MCP server that lets coding agents like Claude directly control ESP32-class hardware, enabling autonomous firmware development, over-the-air deployment, and remote diagnosis. The tool, which costs $2.51 for an ESP32 board, provides 34 tools over streamable HTTP and supports over 1,000 boards across 40 platforms, with features like live on-chip debugging and local-first operation.

read3 min views1 publishedAug 19, 2026
Show HN: I let an AI agent brick my $2.51 ESP32 on purpose
Image: Michielbdejong (auto-discovered)

docs · quick start · mcp tools · cli · platform · discord

nff is an MCP server that gives coding agents direct control over physical hardware — on the bench during development, and in the field for maintenance and diagnosis.

Connect your board over USB and Claude writes, compiles, flashes, and reads serial output autonomously. Deploy devices with the nff-sdk-c

library and Claude can reach them remotely: capture crash state, diagnose failures, and push fixes — without physical access.

nff is the open-source bench CLI of the— an end-to-end, agent-driven system for developing, shipping, and operating ESP32-class firmware (bench → OTA → fleet diagnosis). This repo and the device library ([nff platform]nff-sdk-c

) are the twoMIT-licensedpieces that run on your laptop and hardware; the hosted backend is proprietary.

you: "Run the sensor init sequence and assert the calibration values over serial"
LLM: [writes firmware] → [compiles] → [flashes ESP32] → [reads serial] → returns structured output

you: "Why did the unit in the field just hard-fault?"
LLM: [captures panic over OTA] → [reads registers + backtrace] → "Stack overflow in your sensor ISR at line 47"

the bench loop, in one conversation— no switching between editor, terminal, and serial monitor. The agent iterates on firmware in response to serial output, catches exceptions, and reflashes.field maintenance when the firmware is dead— a crashed bare-metal MCU has no shell, no SSH, no process table. nff captures registers, stack, memory, and backtrace and routes them to a cloud agent that explains the failure and drives recovery. This is the gap Mender, balena, and similar OTA tools cannot fill: they need a living network client inside the firmware.ship it over the airnff ota deploy

turns the binary you just built into a staged, ECDSA-signed rollout with per-device tracking and automatic rollback;nff fleet --watch

shows it land.ota →board-universal— any of PlatformIO's ~1000+ boards across ~40 platforms (every ESP32 variant, RP2040/Pico, all STM32 families, AVR, SAMD, Teensy, nRF52, Uno R4, RISC-V…), toolchain auto-installed on first build. arduino-cli remains available as a second backend.boards →live on-chip debugging— real breakpoints, call stacks, and variable inspection over JTAG/SWD (OpenOCD + GDB, driven by nff).debug →local-first— compile, flash, monitor, debug, and the MCP tools need no account and never open a browser. Only OTA,repair

, andagent

require a sign-in.one Rust binary— self-contained, no Python runtime, and it self-updates in the background like Claude Code does.

macOS / Linux:

curl -fsSL https://nanoforgeflow.com/install.sh | sh

Windows (PowerShell):

irm https://nanoforgeflow.com/install.ps1 | iex

Then plug in your board and run:

nff init      # detects the board, writes config, registers + starts the MCP server
nff doctor    # verify

Restart Claude Code so it picks up the MCP server, then just describe what you want:

you: "Flash sketches/blink_esp32 and confirm the LED is toggling over serial"
LLM: [compiles] → [flashes ESP32] → [reads serial] → "LED toggling at 1 Hz, confirmed"

Full install options, --cloud

sign-in, and first-run detail: quick start →

34 tools over streamable HTTP on 127.0.0.1:3010/mcp

, started in the background by nff init

.

Group Tools Covers
Bench
7 list_devices , compile , flash , serial_read /write , reset_device , get_device_info
Debug
14 breakpoints, call stack, variables, registers, memory, stepping, raw GDB
Field
8 diagnose (local, no login), repair (cloud, ELF-symbolized) + auth lifecycle
Fleet & OTA
5 ota_deploy , ota_status , ota_deployments , ota_devices , fleet_status

Full signatures and return shapes: mcp tools →

everything lives at ** nanoforgeflow.com/docs**:

quick start·

cli reference·

configuration·

mcp tools·

using claude code·

device sdk·

provisioning·

ota deploys·

git-push deploys·

fleet status·

crash diagnosis·

on-chip debug·

power·

security

In-repo reference (docs/): boards & USB ids · self-update & config · roadmap · architecture

Bugs and feature requests go to GitHub Issues; read CONTRIBUTING.md before opening a PR — adding a board is usually a two-line change. Please follow the Code of Conduct, and report vulnerabilities via SECURITY.md. Questions and ideas are welcome on Discord.

MIT — see LICENSE.

Copyright (c) 2026 Gauthier Lechevalier

── more in #ai-tools 4 stories · sorted by recency
── more on @nanoforgeflow 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/show-hn-i-let-an-ai-…] indexed:0 read:3min 2026-08-19 ·