Turn Your LAN Into a Codex Powerhouse with Codex LAN Bridge
I built Codex LAN Bridge to make Codex actually useful across multiple machines instead of being stuck on just one box. It lets one Codex installation call another over the local network, so my main session can discover remote agents, send prompts to another machine, start long-running jobs, check progress later, and continue the same remote conversation by reusing the thread ID. Under the hood, it uses a broker, per-machine endpoints, and a local MCP bridge, and the remote side talks to the machine’s real codex mcp-server—so this is built for real work, not a toy demo.
The practical upside is huge. I can stay on the machine I’m actively using while offloading work to an Ubuntu or Windows system that already has the right repo, toolchain, environment, or mapped shares. It supports fast synchronous requests, async jobs for heavier tasks, machine-specific instruction files, health checks with doctor.py, Windows startup automation, and even optional SMB drive mapping before the Windows endpoint launches.
So instead of treating Codex like a single-machine assistant, this turns it into a network-aware workflow tool. The broker can live on Ubuntu, endpoints can run on Ubuntu and Windows, and once the local MCP bridge is installed, those remote machines show up as tools inside Codex. It’s designed for trusted LAN setups, and the repo notes that SSH tunneling or a VPN is the better path if the network isn’t fully trusted.
One thing to note is that files don’t transfer well over the bridge yet. What codex tries to do is zip the files, then batch segments of base64 text. Ewwww. Another way codex tries moves files with the bridge is by creating a temporary ssh/scp connection, which has worked, I’m just not sure how comfortable we all are with that.
I hope this is adopted by the team and improved upon.