Workflow
Shared Memory Between Cursor and Claude Code
Use Origin as the local bridge when planning, editing, and debugging happen in different AI coding tools.
Article packet
Workflows
Developers switching between Cursor and Claude Code
5 min read
01
One local daemon prevents tool-specific memory silos.
02
Claude Code gets the plugin workflow; Cursor gets MCP tools.
03
Both can recall the same decisions and handoffs.
01
Quick answer
Connect both Cursor and Claude Code to the same Origin daemon. Claude Code can use the plugin workflow, while Cursor uses MCP tools; both read and write the same local memory.
Origin makes the daemon the shared boundary. The interface changes by client, but memory, pages, spaces, and git history stay in one local place.
02
When this problem appears
Many developers plan in one tool and edit in another. If each tool has separate memory, the user becomes the synchronization layer.
03
Set up both clients
Use the richest path for each client while keeping one daemon.
- Install the Claude Code plugin and run /init.
- After /init succeeds, run ~/.origin/bin/origin mcp add cursor, or origin mcp add cursor if ~/.origin/bin is on PATH.
- Verify capture in one client and recall in the other with the same daemon, data dir, and space.
- Use the same space for the same project; Cursor should pass the same space in Origin MCP calls or lock ORIGIN_SPACE in ~/.cursor/mcp.json.
- Use /handoff in Claude Code when ending work that Cursor will continue.
Cursor and Claude Code smoke test
/init
~/.origin/bin/origin mcp add cursor
# Capture a harmless fact in Cursor with the same Origin space.
# Recall that fact from Claude Code with /recall.
# Then capture in Claude Code and recall from Cursor.04
What to check next
If the tools appear to disagree, check space selection, daemon URL, and data directory before assuming memory is missing. The prebuilt runtime supports macOS, Linux glibc, and Windows; Claude Code on Windows may involve WSL or Git Bash caveats.
Try the local memory loop
Install Origin, connect your AI client, and verify that capture, recall, and handoff work on your machine.
FAQ