Running Tesla Model 3 Computer on Desk with Salvaged Parts

Hardware Setup from Salvaged Parts
A researcher needed Tesla hardware for bug bounty research and sourced components from crashed vehicles. The Tesla Model 3 computer consists of two parts: the MCU (Media Control Unit) and the autopilot computer (AP) layered together. The unit is iPad-sized, about as thick as a 500-page book, and comes in a water-cooled metal casing.
Searching for "Tesla Model 3 MCU" on eBay revealed multiple listings in the $200-$300 range from salvaging companies that dismantle crashed cars. These sellers often include photos of the original vehicle and allow filtering for parts from the same car.
Required Components
- DC power supply: Adjustable 0-30V model, 10A version recommended (the setup can consume up to 8A at peak)
- Touchscreen module: From salvaged Model 3, found for $175 on eBay
- Display cable: 6-pin cable with Rosenberger 99K10D-1D5A5-D connector (2 pins for 12V/ground, 4 for data)
The cable proved difficult to source individually. Tesla publishes wiring "Electrical Reference" documents publicly on their service website, showing exact pin configurations. The researcher discovered the cable is similar to LVDS cables used in BMW vehicles and ordered a compatible version.
Powering Up and Network Access
Using Tesla schematics and online references, the researcher identified the correct pins for 12V and ground on the computer's cut cables. After attaching power supply clips to the stripped wires, red LEDs began flashing and the computer started.
Without a screen, interaction was limited. The researcher connected an Ethernet cable to the port next to the power connector. The car's internal network lacks DHCP, requiring manual IP configuration: 192.168.90.X/24 with X higher than 105 to avoid conflicts.
Internal Network Details
From an older /etc/hosts file found on Reddit:
- 192.168.90.100 cid ice # mcu
- 192.168.90.102 gw # gateway
- 192.168.90.103 ap ape # autopilot
- 192.168.90.105 ap-b ape-b # also autopilot
- 192.168.90.60 modem # has the ftp server
SSH on port 22 and a webserver on port 8080 were open on 192.168.90.100 (the MCU), confirmed to still be accessible on newer models.
📖 Read the full source: HN AI Agents
👀 See Also

Managing Context Limits in Long Claude Runs: AC Tree Pattern
A developer shares a failure pattern in long Claude runs where auto-compact causes information loss and context limits prevent continuation, then describes a solution using an AC tree dependency graph with isolated sessions per node.

Running Gemma 4 as a Local Autonomous Agent with Claude Code on 16GB VRAM
A developer successfully configured Google's Gemma 4 31B model to function as a local autonomous coding agent through Claude Code CLI v2.1.92, overcoming VRAM limitations and parsing issues using llama.cpp b8672 and custom Python routing.

Automating Recruiting Workflows with Claude Desktop: A Case Study
A developer automated the first layer of recruiting using Claude Desktop, Chrome with browser extension, and Google Calendar integration, handling resume screening and interview scheduling every two hours on a Windows workstation.

How Claude Turned a Non-Developer's Site into 10K Users with SEO and AEO
A non-developer used Claude for SEO content strategy, AEO optimization, and technical audits to grow an AI skill marketplace from 0 to 10,000 active users in 6 weeks with $0 ad spend.