Our team pays ~$700/month for Claude Code Max (3 accounts). We're Claude-native. But between Claude's recent reliability issues and Codex's cost advantages for simpler workloads, we've been moving some work to Codex.
Migrating turned out to be annoying — plugins, MCP servers, memory files, harness configs all live in different places with different formats.
We wrote a CLI to automate it and open sourced it.
📚 Full writeup: https://treesoop.com/blog/claude2codex-migration-tool-open-source-2026
🔧 GitHub: https://github.com/treesoop/claude2codex
What it migrates
| Claude Code | → | Codex |
|---|---|---|
~/.claude/CLAUDE.md |
~/.codex/config.md (format converted) |
|
~/.claude/settings.json |
codex.toml |
|
~/.claude/skills/*.md |
~/.codex/prompts/*.md |
|
~/.claude/user_profile.md |
Codex profile | |
| MCP server registrations | Codex-compatible config block | |
| Harness trigger logic | Best-effort port with warnings |
Install
npx claude2codex init
npx claude2codex migrate --dry-run # preview
npx claude2codex migrate # execute
Who should care
- Teams running both Claude Code and Codex in hybrid mode
- Anyone hitting Claude Code session limits or reliability issues
- Teams evaluating Codex but not wanting to set up from scratch
Results in our team
Nine team members migrated their setups using this tool. ~95% of settings auto-converted and worked in Codex on first try. The remaining 5% were flagged in a conflict report and manually adjusted.
Why hybrid, not replacement
We still default to Claude Code Max for anything requiring strong reasoning or long-horizon planning. Codex picks up:
- Short repetitive tasks (doc generation, test writing)
- Tasks where token cost matters more than depth
- Fallback when Claude adaptive thinking underallocates
Details
- MIT licensed
- Your original Claude config is preserved (not modified)
- Codex unsupported features get warnings in a report
More from TreeSoop: ai-news-mcp, hwp-mcp, whisper_transcription
