Sniff and reverse-engineer a car's LIN bus: capture, changing bits, speed detect, LDF decode
The io.github.Lukasz-Juranek/lin-usb-converter MCP server focuses on sniffing and reverse-engineering a car’s LIN bus. It captures LIN bus traffic, can change bits, detects speed, and decodes LDF to interpret message definitions.
🛠️ Key Features
Sniff and reverse-engineer a car’s LIN bus
Capture bus traffic
Change bits on the bus
Detect speed
Decode LDF
🚀 Use Cases
Analyzing how LIN messages behave on a vehicle network
Troubleshooting or experimenting with LIN communications
Converting or interpreting LDF-based message information
Studying speed-related signals on the LIN bus
⚡ Developer Benefits
Provides tooling-oriented access to LIN bus capture and LDF decode
Open source USB ↔ LIN bus dongle. LIN 2.1 master, slave or monitor. SLCAN-compatible virtual COM port - works on Windows, Linux and macOS with no proprietary drivers.
USB LIN Converter (uLC) - dongle photo
The USB LIN Converter (uLC) is a small open-source USB dongle (STM32F042, ~31 × 18 mm) that lets a PC act as a LIN bus master, slave or passive monitor. It enumerates as a USB CDC virtual COM port and speaks an SLCAN-style ASCII protocol, so it bridges into Linux SocketCAN via slcand and works with parts of can-utils.
A commercial automotive LIN analyzer typically costs hundreds to thousands of euros. This one is around $27 on Tindie.
The LIN protocol implementation is based on openLIN.
Features
LIN 2.1 support
Monitor mode - full frame reception plus single-frame transmission
Master mode - up to 16 scheduled frame slots
Slave mode - up to 16 scheduled frame slots for slave simulation
9600 / 19200 baud rate selection
Optical isolation between LIN and the USB / PC side
Software-controlled termination resistors
External 4.5 - 28 V powers the LIN transceiver
Device size 31 × 18 mm (without USB connector)
USB DFU bootloader for firmware updates
No proprietary drivers - Linux / macOS / Windows
MCP server so AI agents can sniff and reverse-engineer a LIN bus (see below)
Plug the device into a USB port. It enumerates as a virtual COM port (COM3 on Windows, /dev/ttyACM0 on Linux, /dev/cu.usbmodem* on macOS).
No driver is needed on Windows 10/11, Linux or macOS: the built-in USB CDC driver is used. Since firmware 3.5 the device has its own USB ID from pid.codes, 1209:2779; older firmware used ST's generic 0483:5740 (on Windows 7/8 that needs the ST Virtual COM Port driver).
Open the port at any baud rate (USB CDC ignores baud). Line ending must be CR (0x0D).
Power the LIN side with 4.5 - 28 V on the external supply pin.
Send S0 to select 19200 baud, then l to enter monitor mode. Bus traffic will start to appear.
Linux SocketCAN bridge
bash
sudo slcand -o -s8 -t hw -S 921600 /dev/ttyACM0 slcan0
sudo ip linkset slcan0 up
candump slcan0
Python (pyserial) sniff snippet
python
import serial
ser = serial.Serial('/dev/ttyACM0', timeout=1)
ser.write(b'S0\r') # 19200 baud
ser.write(b'l\r') # monitor modewhileTrue:
line = ser.readline()
if line:
print(line.decode(errors='replace').strip())
A higher-level Python library is provided in python_lib/ (package ucanlintools).
AI agents (MCP server)
python_lib/ucanlintools/mcp_server.py is an
MCP server that lets an AI agent sniff and reverse-engineer a LIN bus: capture traffic
with per-id periods and changing bits, log every data change while you press a button,
find the bus speed and checksum type automatically, and decode or encode signals from
an LDF file. It is read-only by default; sending frames and acting as master or
slave needs --allow-transmit.
bash
uvx ucan-lin-mcp --mock # from PyPI, no install step (Python 3.10+)# or from this repo: pip install -e "python_lib[mcp]" && ucanlin-mcp --mock
Register it with any MCP client as a stdio server:
--mock runs a simulated bus. mcp_package/ is the ucan-lin-mcp launcher
package and server.json its MCP Registry entry.
Tools: device_info, monitor_bus, detect_bus, load_ldf, decode_frame,
encode_frame, and with --allow-transmit also send_frame, run_master, run_slave.
ASCII command reference
All commands are ASCII terminated with CR (0x0D). Errors return 0x07 (BEL).
Command
Description
O
Open in master mode
L
Open in slave mode
l
Open in monitor mode
C
Close LIN port
Sx
Baud: S2 = 9600, others = 19200
V / v / N
Hardware / firmware version / serial number query
tiiildd...
Transmit full header + data frame (id, length, data)
r0iil
Header only request (slave answers)
r1ff0
Master mode: start sending frames from schedule table
r2ff0
Clear schedule table
r31 / r30
Checksum for all frames: r31 classic (LIN 1.x), r30 enhanced (LIN 2.x, default). Diagnostic frames 0x3C/0x3D always use classic. Replies Z. Kept until power off; t31/t30 work too.
t04bbbb
Custom LIN baud rate, bbbb = 4 hex digits (e.g. t0428B1 = 10417 baud). Send before adding frames. Replies Z.
T0iipptt0ldd... / R0iipptt0l
Frame with explicit delay pp and response wait time tt (ms, hex), e.g. T00A643001A5. pp is the time from the start of the previous slot's header to this slot's header (firmware 3.4 and later; older firmware started it when the previous frame ended).
Commands are executed one at a time: wait for the reply (CR, z/Z or BEL) before sending the next one.
LIN 2.2A specification, MCU datasheet, LIN transceiver datasheet and supporting documents.
Build
Source is pure C. Import the Eclipse project in embedded/TrueSTUDIO/LUCEmbedded into STM32CubeIDE (tested with 1.19) and build the Debug configuration; the image is Debug/LUCEmbedded.bin. The openLIN stack is vendored in embedded/Src (open_lin_*.c/h, implementation/ST32). The USB CDC layer was generated with STM32CubeMX from embedded/LUCEmbedded.ioc.
Bootloader
The STM32F042 has an embedded ROM USB DFU bootloader. To enter it from a running device send boot\r over the virtual COM port. See this tutorial for the flashing flow.
With STM32CubeProgrammer installed, the flashing can be scripted:
LDF parsing, signal encode/decode, frame formatting
test_luc_hardware.py
1-2 dongles
Library: version query, mode switching, monitor loopback, master → slave
test_firmware_protocol.py
1-2 dongles
Firmware over raw serial: malformed-command rejection, 16-row table limit, all frame lengths 1-8 in both directions with exact data check, live data update, diagnostic frames 0x3C/0x3D, unanswered headers, extended T/R row timing, custom baud rate, monitor reception (including back-to-back frames) and monitor transmit. Parametrized for 19200 and 9600 baud.
test_lib_simulation.py
2 dongles
Scripted MasterSimulation.py / SlaveSimulation.py: LDF-encoded master frame, slave response, both changing at runtime, at both speeds
Setup: two converters on one LIN bus (LIN pins connected, external LIN supply applied). LUC_MASTER_PORT is used as LIN master, LUC_SLAVE_PORT as slave / monitor. Tests skip automatically when a port can't be opened, so the offline tests always run.
bash
cd python_lib
pip install pyserial lark pytest
LUC_MASTER_PORT=COM3 LUC_SLAVE_PORT=COM56 PYTHONPATH=. pytest tests -v
# swap the ports and run again to test each converter in both roles
LUC_MASTER_PORT=COM56 LUC_SLAVE_PORT=COM3 PYTHONPATH=. pytest tests -v
On Windows PowerShell set the variables with $env:LUC_MASTER_PORT="COM3" etc.
How firmware 3.8 was tested
Same setup, both converters flashed with LUC_0308.bin.
Monitor fix: frames sent back to back (each one as soon as the previous command was acknowledged, about 4 ms apart at 19200 baud) leave less idle time than the monitor's end-of-frame timeout, so the next break reset the reception and the previous frame was lost. The monitor now reports the previous frame when the next break arrives. Before: 15 of 32 frames received; with the fix: 32 of 32, in order.
New regression test test_monitor_reports_back_to_back_frames (fails on 3.7, passes on 3.8); full suite run twice with the roles swapped: 66/66 passed in each direction.
Same setup, both converters flashed with LUC_0307.bin.
Slave fix traced byte by byte with a diagnostic build: when the slave subscribed a frame nobody answered, the next header's break byte was counted as that frame's data, and the stale count made the slave take the following frame's last data byte as its checksum and drop it. With the fix the slave receives every following frame (23 of 23 in 1.5 s, before 0).
New regression test test_slave_subscribed_to_unanswered_header_keeps_next_frame; full suite run twice with the roles swapped: 65/65 passed in each direction.
How firmware 3.6 was tested
Same setup, both converters flashed with LUC_0306.bin.
Clock recovery checked with a test-only build that reported the CRS registers: the CRS synchronised to the USB SOF within the first seconds (SYNCOK set, no errors afterwards) and moved the HSI48 trim from 32 to 35 on one converter, i.e. that converter's clock ran measurably slow without it.
Full suite run twice with the roles swapped: 63/63 passed in each direction; master slot periods unchanged from 3.4 (within about 1 ms of the configured delay).
How firmware 3.5 was tested
Same setup, both converters flashed with LUC_0305.bin. They enumerate as 1209:2779 and Windows 10 loads its built-in CDC driver without any installation.
Full suite run twice with the roles swapped: 55/55 passed in each direction. Firmware 3.5 changes only the USB ID, the LIN code is the same as 3.4.
How firmware 3.4 was tested
Two uLC converters (hardware V0001), both flashed with LUC_0304.bin, connected on one powered LIN bus.
Full suite run twice with the roles swapped: 55/55 passed in each direction (110 test runs).
Master slot timing measured with the second converter as monitor (3 s window, 19200 baud): the slot period now matches the configured delay within about 1 ms for 0, 2, 4 and 8 data bytes, for 10, 20 and 40 ms delays, and for unanswered subscriber slots. Firmware 3.3 added the frame time to every slot (+2 to +6 ms) and the full response timeout to unanswered subscriber slots (+12 ms).
How firmware 3.3 was tested
Two uLC converters (hardware V0001), both flashed with LUC_0303.bin, connected on one powered LIN bus.
Full suite run twice with the roles swapped: 47/47 passed in each direction (94 test runs).
Bus throughput measured during testing (4 s window, 16-frame schedule): every frame scheduled 12-14 times at 19200 baud and 10-11 times at 9600 baud, no lost or corrupted frames, no unexpected serial output.
The bugs fixed in 3.3 each have a regression test: malformed row locking the table, slave mode stalling after the first response at 9600 baud, slave mode not answering at all, master not reporting slave responses, 6-8 byte responses timing out at 9600 baud, and the classic checksum setting being ignored. The checksum tests were also run against the previous firmware and fail there, as expected.
FAQ
What is the cheapest USB LIN bus converter that works on Linux?
The uCanDevices USB LIN Converter (uLC) is an open source dongle around $27 that enumerates as a USB CDC virtual COM port on Linux, macOS and Windows. It uses an SLCAN-style ASCII protocol so it bridges into SocketCAN via slcand.
Does it support master and slave mode?
Yes - monitor (full reception plus single-frame transmission), master (up to 16 scheduled slots) and slave (up to 16 slots for slave simulation). Modes are selected with the O, L and l commands.
Is the LIN side optically isolated from the USB host?
Yes. The LIN transceiver is optically isolated from the PC side, protecting the host from automotive transients. The LIN side needs an external 4.5 - 28 V supply since USB only powers the MCU.
Which LIN baud rates and LIN version are supported?
LIN 2.1, with selectable 9600 and 19200 baud. Classic and enhanced checksum are switchable with the r31 / r30 commands (firmware 3.3 and later). Custom frame timing is supported in master mode.
Can I use it to flash an ECU over LIN?
It is a generic LIN interface. ECU flashing depends entirely on the target ECU's own LIN bootloader; if that bootloader is documented or already supported by an existing tool, you can drive it through this dongle.
Is the firmware open source?
Yes - this repository contains firmware (embedded/), hardware (pcb/) and the Python helper library (python_lib/). The device also has a USB DFU bootloader for in-field updates.