Codex CLI release rust-v0.160.1, dated October 5, 2026, now preserves SYSTEMROOT, TEMP, and TMP when the tool launches remote stdio MCP servers that specify explicit remote environment variables. Those three variables are not cosmetic. SYSTEMROOT tells Windows where its core system files live, and TEMP and TMP control where processes write temporary data. Before this fix, specifying remote environment variables for an MCP connection apparently dropped these defaults, which on Windows can break process launches or file operations that assume they exist.
The decision this release actually changes is whether teams running Codex CLI on Windows can safely configure remote MCP servers with custom environment variables at all. Before 0.160.1, doing so risked an incomplete environment, a failure mode that would not necessarily announce itself clearly and could instead show up as an obscure tool crash or write error. Teams that avoided setting explicit remote environment variables on Windows specifically to dodge this were working around a bug they may not have known the shape of. OpenAI's release notes do not disclose how long the gap existed or how many Windows deployments were affected before the patch.
For procurement and platform teams, the fix is a reminder that cross-platform parity in developer tools is not automatic, and Windows-specific regressions can sit unnoticed in a stack built primarily by and for Unix-like environments. Anyone who standardized on Codex CLI for Windows-based engineering environments should check whether prior workarounds, like avoiding remote env var overrides or hardcoding TEMP paths elsewhere, are still in place and now redundant or conflicting. There is also an open question the release notes do not answer: whether versions before 0.160.1 are still receiving security patches, which matters for teams that cannot upgrade immediately.
None of this changes pricing, licensing, or contractual lock-in, since Codex CLI's release cadence is a product maintenance matter, not a commercial one. But it does affect switching cost calculations in a narrower sense: teams evaluating whether to build Windows-specific tooling around Codex CLI's MCP integration now have one fewer silent failure mode to engineer around, which lowers the operational cost of staying rather than migrating to a competing agent tool with its own undocumented platform quirks. Buyers should treat this as evidence that platform-specific bugs in this category of tool are real and worth testing for directly, not just trusting release notes.
The fix is narrow, it is dated, and it only covers three variables, which is itself a small admission that the list of what Windows needs preserved was incomplete until now.