Workflow
Shared Memory Between Codex and Claude Code
Use Origin to carry implementation context between Codex sessions and Claude Code plugin workflows.
Article packet
Workflows
Developers using both Codex and Claude Code
5 min read
01
Codex connects through MCP-only setup.
02
Claude Code gets plugin commands and the same daemon.
03
Handoffs make cross-tool continuation explicit.
01
Quick answer
Use Origin as the shared store: configure Codex through MCP, configure Claude Code through the plugin, and verify that a capture from one can be recalled from the other with the same daemon, data dir, and space.
Origin keeps decisions, lessons, pages, and handoffs independent of the tool that produced them. That lets Codex and Claude Code participate in one work history.
02
When this problem appears
Codex may handle one coding pass while Claude Code handles another. Codex native Memories and AGENTS.md still help Codex itself, but they do not automatically become Claude Code memory.
03
Verify cross-tool memory
Do a small round trip before trusting the workflow.
- Set up Origin once.
- Add Codex with ~/.origin/bin/origin mcp add codex.
- Install the Claude Code plugin with /plugin marketplace add 7xuanlu/origin, /plugin install origin@7xuanlu, then /init.
- Capture a harmless project fact in Codex with space X.
- Recall that fact from Claude Code with the same space.
- Reverse the smoke test: capture in Claude Code, then recall from Codex.
Codex and Claude Code smoke test
~/.origin/bin/origin mcp add codex
/plugin marketplace add 7xuanlu/origin
/plugin install origin@7xuanlu
/init
# Use the same Origin space for both clients during capture and recall.04
What to check next
Do not use different daemons, data directories, or spaces unless you intentionally want isolated memory stores. MCP-only Claude Code is tools-only; the plugin path is what provides /init, /brief, and /handoff.
Try the local memory loop
Install Origin, connect your AI client, and verify that capture, recall, and handoff work on your machine.
FAQ