A 1994 Jeep Cherokee has no OBD2 port and no scan tool that works on it. Its owner, Muhammad Zulali, wired a $3 Arduino Nano to the diagnostic plug under…
// the short version
A 1994 Jeep Cherokee has no OBD2 port and no scan tool that works on it. Its owner, Muhammad Zulali, wired a $3 Arduino Nano to the diagnostic plug under the hood and had Claude Code and Codex work out how the engine computer (Chrysler SBEC, OBD1) talks.
// what to take away
- A 1994 Jeep Cherokee has no OBD2 port. Its owner, Muhammad Zulali, wired a $3 Arduino Nano to the diagnostic plug and had Claude Code and Codex work out how the engine computer (Chrysler SBEC, OBD1) talks over SCI, the dealer scan tool's line: 7812.5 baud, electrically inverted, and every byte echoed before the next one is sent.
- Reading a value is a short exchange: send 14 and a parameter ID, wait for both echoes, and the computer repeats the value byte until it gets 00. Battery voltage C6 is 198 x 0.0625 V = 12.38 V, and a row of about 12 parameters takes roughly half a second.
- ChryslerScanner's formula treats the top bit of long-term fuel trim (0E) as a sign, so raw 110 read as +21.6 %. The computer's own code says 128 is no correction and each count is 1/384 of the fuel, so 110 is -4.7 %. The misread sent a day of testing after a fuel problem that didn't exist. Built and tested on one 1994 Cherokee.
What xj-scanner is, and what it is not
xj-scanner is an open-source (GPL-3.0) project by Muhammad Zulali that reads the engine computer of a 1991 to 1995 Jeep from a laptop. Cars of that age predate OBD2, so there is no standard port and no off-the-shelf scan tool for them. The project's answer is three wires from the diagnostic plug under the hood to an Arduino Nano clone (about $3), and the Nano plugged into a laptop over USB. A Python scanner on the laptop then reads fault codes and live engine data and logs drives to CSV files in pc/logs/ (AGENTS.md).
It was built and tested on exactly one car: a 1994 Jeep Cherokee XJ with the 4.0L engine and engine computer part number 56030117. The repository says other 1991 to 1995 Chrysler and Jeep computers of the same family (SBEC, OBD1) should be close, and tells owners to measure their own plug before connecting anything. The wiring and those checks live in the repository's docs; this page does not repeat them.
The repository is also written to be driven by a coding agent. Its AGENTS.md is a playbook the agent follows to walk a person through setup one step at a time, and the owner did the work with Claude Code and Codex, as his video describes.
How the computer talks: SCI, one echoed byte at a time
The diagnostic line on these computers is SCI, the serial line the dealer's DRB scan tool used. Per docs/protocol.md:
- 7812.5 baud, 8 data bits, no parity, 1 stop bit.
- Electrically inverted compared with a normal serial port: with the key on the line idles low, and a start bit is high. The Nano handles this in software (SoftwareSerial with inverse logic), so there is no inverter chip.
- Two data wires plus ground. The computer listens on one and answers on the other.
- Every byte you send is echoed back. The tools send one byte, wait for its echo, then send the next. Bursts get dropped.
- Some computers swap the two 4-bit halves of each byte. The scanner checks for that by testing whether the battery voltage reads sanely; the test car uses normal order.
The Nano runs a small bridge sketch. Over USB at 115200 baud it answers PROBE (how often the computer's transmit line read high, near 0 when connected with the key on) and GO I or GO N, after which it passes bytes straight through in inverted or normal logic.
Reading one parameter, byte by byte
Live data is read with command 14 followed by a parameter ID. The battery voltage exchange from the docs:
laptop -> 14 computer -> 14 (echo)
laptop -> 0A computer -> 0A (echo)
computer -> C6 (the value)
computer -> C6 C6 (repeats)
laptop -> 00 (stop)0xC6 is 198, and battery voltage scales at 0.0625 V per count, so 198 x 0.0625 = 12.38 V. The computer keeps repeating the value byte until it receives 00. A full row of about 12 parameters takes roughly half a second; a mode that reads only the O2 sensor manages about 20 readings a second. Fault codes are read with 10, which returns one byte per stored code, end markers and a checksum (10 FE 0E means no codes).
The twelve parameters it logs, and how each byte scales
By default the scanner logs engine speed (raw x 32 rpm), coolant and intake air temperature (1.8 x (raw - 128) + 32 F), throttle position (raw x 0.0196 V), manifold pressure (raw x 0.059756 psi), the upstream O2 sensor (raw x 0.0196 V), battery voltage (raw x 0.0625 V), spark advance (raw x 0.5 degrees), target idle air position (raw steps), vehicle speed, long-term fuel trim, and fuel system status. The names and formulas come from ChryslerScanner by László Dániel unless the docs say otherwise.
Fuel system status (3B) does not answer on this computer, and neither does 57, so there is no "closed loop" flag. Closed loop shows up instead as the O2 voltage switching back and forth across about 0.45 V.
The fuel trim byte, read two ways
Long-term fuel trim (0E) is the correction the computer has learned to apply to the fuel it injects. The repository's docs list it first among the "quirks that will mislead you":
- 128 means no correction. Below 128 the computer is removing fuel; above 128 it is adding.
- One count is 1/384 of the fuel, about 0.26 %. The multiplier is
(256 + raw) / 384, so 128 gives exactly 1.0. - The computer keeps it between 108 and 168, which is -5.2 % to +10.4 %.
- ChryslerScanner's formula treats the top bit as a sign. Raw 110 comes out as +21.6 % under that formula. Under the computer's own scale it is 18 counts below 128: (256 + 110) / 384 = 0.953, a correction of -4.7 %.
On the test car the wrong +22 % "sent a whole day of testing after a fuel-delivery problem that didn't exist", in the docs' words. The question was settled from the computer's own code: the routine that applies the trim and the reset command that writes 128. The repository's README credits the published disassembly it was settled from. The scanner now logs the trim as counts from 128 (so -18 for raw 110); multiply by 0.26 for a percentage.
Two consequences are worth spelling out, and the arithmetic is ours. First, a reading of +21.6 % is outside the range the computer allows at all (the ceiling is +10.4 %), so the formula's answer could never have been a value the trim actually held. Second, the formula did not only put zero in the wrong place: 21.6 % over 110 counts is about 0.196 % per count, while the computer's own step is 1/384 of the fuel, about 0.26 %. The zero and the step size were both wrong.
The other quirk: vehicle speed reads double
On the test car the vehicle speed parameter (41) reads about twice the real speed (it showed 56 on a residential street). The repository's drive report takes a scale factor to halve it, and the docs tell owners to compare it with their speedometer. Like the fuel trim, it is a reminder that a scaling table copied from another tool is a hypothesis until the car agrees.
Reading is safe; some commands are not just reading
The agent playbook draws a clear line. Reading fault codes and live data is safe. Three things are not just reading, and the agent is told to ask before each and say what it does: erasing fault codes (which also clears what the computer has learned about fuel), resetting the learned fuel trim, and actuator tests that switch real hardware. The playbook also has the agent prove the laptop side first against a simulated Jeep, so that any later failure can only be wiring, and tells it not to guess at a diagnosis from a reading that may be a scaling quirk. This video covers only the reading.
Terms
- OBD1
- The older, manufacturer-specific on-board diagnostics that came before OBD2; this Jeep's computer is OBD1.
- OBD2
- The standard diagnostic port and protocol on later cars, which modern scan tools expect; this 1994 Jeep does not have one.
- SBEC
- The family of Chrysler and Jeep engine computers (OBD1) used on 1991 to 1995 models like this one.
- PCM
- What the repository's docs call the engine computer (the ECU); the test car's is part number 56030117.
- SCI
- Chrysler's serial communications interface, the diagnostic line the dealer's DRB scan tool used; 7812.5 baud and electrically inverted on these computers.
- Echo
- The computer repeats every byte it receives, which lets the sender confirm each byte before sending the next.
- Long-term fuel trim
- A correction the engine computer learns over time and applies to the fuel it injects; on this computer 128 is neutral and each count is 1/384 of the fuel.
- ChryslerScanner
- An open-source scan tool for Chrysler vehicles by László Dániel, the source of xj-scanner's parameter IDs and scaling formulas.
- Sign bit
- Treating the top bit of a byte as plus or minus, which makes 0 the middle of the range instead of 128.
- AGENTS.md
- A file in a repository addressed to coding agents; here, the step-by-step playbook an agent follows to help someone connect and read the car.
Questions
Can a 1994 Jeep be read without an OBD2 port?
Yes, over its OBD1 diagnostic plug. xj-scanner does it with an Arduino Nano bridging the computer's SCI line to a laptop, tested on one 1994 Cherokee XJ 4.0L.
Why did the scan-tool formula read the fuel trim as +21.6 %?
ChryslerScanner's formula treats the top bit of the byte as a sign, so zero is its middle and raw 110 reads as a large positive correction, with a step of about 0.196 % per count. The computer's own code centres the trim on 128, which makes 110 a correction of -4.7 %.
What does a long-term fuel trim of 110 mean on a Jeep SBEC computer?
110 is 18 counts below the neutral 128. At 1/384 of the fuel per count, the computer is removing about 4.7 % of the fuel.
Did Claude Code do this on its own?
No. The owner wired the Nano to his car and worked with Claude Code and Codex over about a week, by his own account. The repository's AGENTS.md is written so an agent can guide a person through the same steps.
Will this work on my car?
It was built and tested on one car. The repository says other 1991 to 1995 Chrysler and Jeep SBEC computers should be close, and that you should measure your own plug before connecting anything.
// further reading
- MZULALI/xj-scanner ↗: the repository, with the protocol, the parameter table and the wiring docs
- Claude reverse engineered my 32-year-old Jeep ↗: Muhammad Zulali's own video of the build, 10 November 2026
- ChryslerScanner ↗: László Dániel's scan tool, the source of the parameter tables
// transcript
One byte from a 1994 Jeep got read the wrong way round. The scan tool's formula said 22 % extra fuel, when the computer was really taking some away, and that wrong number sent a whole day of testing after a fuel problem that didn't exist. It's a Cherokee with no OBD2 port, and no scan tool that works on it. So its owner, Muhammad Zulali, wired a $3 Arduino to the diagnostic plug and had Claude Code and Codex work out the rest. The computer talks over SCI, the serial line the dealer's scan tool used, and it echoes every byte, so you send one, wait for the echo, then send the next. Ask for the battery with 14 0A, and it answers C6, over and over, until you send 00. That's 198 sixteenths of a volt, 12.38 volts. About a dozen readings take half a second, enough to log a whole drive. What each byte means came from ChryslerScanner, a community tool by László Dániel, including long-term fuel trim, the correction the computer has learned to add to or take from the fuel. On this Jeep it read 110. ChryslerScanner treats the top bit as a sign, with zero in the middle, so 110 comes out as +21.6 %, which points straight at a fuel delivery problem. The computer's own code settles it: the routine that applies the trim, and the reset, which writes 128. 128 is the middle, no correction at all, and each step is 1/384 of the fuel, about a quarter of a percent. The trim can't even get near that formula's number, because the computer caps it at +10.4 %. So 110 is eighteen steps below the middle, -4.7 %, an engine trimming a little fuel away, not one starving for more. The wires, the echoes and the byte were right the whole time, the tool's formula just had the wrong zero, and the wrong size of step.
// source
This explainer is based on xj-scanner: read the engine computer of a 1991-95 Jeep from a laptop by Muhammad Zulali (MZULALI) ↗. The original reporting and technical work belong to its publisher.