BugBuster Gives AI the Tools to Debug Physical Hardware

A few years into the latest AI boom, we finally have a reasonably clear view of what these tools are good at and what is hype. They are unlikely to take over the world anytime soon, but they can give engineers a bit of a productivity boost when used correctly. In particular, software engineers have found them to be helpful when debugging complex code. But these tools are, naturally, limited to assisting with digital tasks. That limits how much help they can be with hardware troubleshooting.

A new open source project called BugBuster is trying to lift that restriction by providing a way for AI models to measure, control, and debug real hardware. The hardware platform, developed by Lorenzo Karavania, is able to interact with popular AI assistants such as Claude through a Model Context Protocol (MCP) server. This allows the AI to measure voltages, drive outputs, capture waveforms, and analyze digital signals as needed to debug hardware issues autonomously.

BugBuster combines the capabilities of several common bench instruments into one platform. The system includes a four-channel 24-bit ADC for measuring voltage, current, and resistance, along with a four-channel DAC capable of generating voltage or current outputs. A waveform engine can create sine, square, triangle, and sawtooth signals, while a bank of level-shifted digital I/O pins and an integrated switching matrix allow signals to be routed dynamically between subsystems.

The mainboard is based on an ESP32-S3 microcontroller with 16 MB of flash and 8 MB of PSRAM. It handles most of the analog measurement and control functions, USB Power Delivery negotiation, I/O routing, and communications. Sitting on top of that is an RP2040-based expansion HAT that adds higher-speed debugging features, including a four-channel logic analyzer capable of streaming captures over USB. The logic analyzer supports sampling rates up to 100 MHz with run-length encoding compression, while a single-channel offline capture mode can reach 125 MHz.

The HAT also incorporates a CMSIS-DAP compatible SWD probe for debugging ARM-based embedded systems. That allows BugBuster to interface directly with tools such as OpenOCD and pyOCD without requiring an additional debug adapter. Adjustable power rails, per-rail current monitoring, and onboard WS2812B status LEDs round out the hardware feature set.

In addition to firmware for both microcontrollers, BugBuster includes a Tauri- and Leptos-based desktop application, a Python control library, an HTTP REST API, MicroPython scripting support, and more than 400 automated tests. The MCP server exposes 59 AI-callable tools, giving compatible assistants the ability to perform tasks such as I2C scans, SPI transfers, logic trace captures, and rail voltage adjustments through natural language requests.

Because handing physical hardware control to an AI system carries obvious risks, BugBuster also implements several hardware and software safety mechanisms. Current limits, e-fuse protection, voltage confirmation requirements, routing restrictions, and automatic post-operation fault checks are all enforced at the tool layer, preventing the AI from bypassing protections directly.

The PCB design files, firmware, and documentation are available in the project’s GitHub repository. Check it out if you’d like to try BugBuster for yourself.