内容
# aidbg
A collection of lightweight debugging tools, each exposing its own MCP server
so an AI can help you debug live hardware/systems over the Model Context
Protocol.
Each tool is a standalone process: you start it, and the AI connects to its MCP
endpoint as a client. The AI does **not** start anything by itself.
## Layout
```
aidbg/
├── xserial/ # serial-port interactive terminal + MCP server
├── xrelay/ # 4-channel USB relay board MCP server
├── xcamera/ # camera MCP server (local preview + photo/frame capture)
└── ... # more tools coming
```
Each subdirectory is self-contained (own `README.md`, own dependencies).
## Hardware
The tools run against a physical debugging rig — the **AIDBG 调试器** (AIDBG
debugger): a fully-automatic debug fixture DIYed from off-the-shelf modules
inside a 3D-printed enclosure. On the outside it only needs an external
12 V DC supply and a single USB Type-C cable to the host.
Internals:
- 12 V → 5 V DC-DC module
- 7-port USB hub — 2 ports are used by the devices below, the rest are free
for the board under test (e.g. a USB-UART adapter for `xserial`)
- 4-channel USB relay board, driven by `xrelay`
- USB camera on a flexible gooseneck arm, driven by `xcamera`
Relay channels:
- 2 channels switch the external 12 V / 5 V power outputs (screw terminals on
the front) to power-cycle the board under test
- 2 channels are reserved for simulating key presses and similar triggers
With `xserial` (serial console) + `xrelay` (power switching / key simulation)
+ `xcamera` (visual feedback) over MCP, the AI can power-cycle a board, watch
it through the camera and drive its console — fully automatic debugging.
| | |
|-------------------------------------------------------------------------------------|-------------------------------------------------------------------------------------|
|  |  |
|  |  |
## MCP endpoints
| Server | Description | Default endpoint |
|----------|----------------------------------------------|--------------------------------|
| `xserial`| Shared serial session + interactive terminal | `http://127.0.0.1:30000/mcp` |
| `xrelay` | 4-channel USB relay board (auto-finds its node) | `http://127.0.0.1:30001/mcp` |
| `xcamera`| Camera capture (photo + frame sequences) | `http://127.0.0.1:30002/mcp` (preview `http://127.0.0.1:30003/`) |
Every server binds `0.0.0.0` by default, so it is reachable both locally and
from other machines on your LAN. From another machine, replace `127.0.0.1`
with this host's LAN IP, e.g. `http://192.168.x.x:30002/mcp`. Start any server
with `--host 127.0.0.1` to restrict it to this machine. There is **no auth**,
so only expose these on networks you trust.
## Connect an AI (opencode.json)
```json
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"xserial": {
"type": "remote",
"url": "http://127.0.0.1:30000/mcp",
"enabled": true
},
"xrelay": {
"type": "remote",
"url": "http://127.0.0.1:30001/mcp",
"enabled": true
},
"xcamera": {
"type": "remote",
"url": "http://127.0.0.1:30002/mcp",
"enabled": true
}
}
}
```
If a service isn't running, opencode can't fetch its tools. Have the AI call
that server's status tool first (e.g. `serial_info`) to confirm it's up; if it
fails, tell the AI to ask you to start the server.
## Getting started
Pick a tool and follow its own README, e.g.:
```bash
cd aidbg
python3 -m venv .venv
.venv/bin/pip install mcp pyserial
.venv/bin/python xserial/xserial.py /dev/ttyUSB0 115200
```
Relay board (auto-finds the USB node; `--device` pins it):
```bash
.venv/bin/python xrelay/xrelay.py
```
Camera debugging (needs `ffmpeg` on PATH). Start it, then open the local
preview to aim the camera at your board:
```bash
.venv/bin/pip install mcp
.venv/bin/python xcamera/xcamera.py # preview: http://127.0.0.1:30003/
```
## License
MIT, see [LICENSE](LICENSE).
MCP 市场