Setup
Cursor Memory MCP: How to Add Local AI Work Memory
Wire Cursor to Wenlan's local MCP memory server so coding sessions can capture and recall project context.
01
Wenlan setup installs the local daemon and MCP connector.
02
wenlan connect cursor writes the Cursor-side MCP configuration when supported.
03
Cursor Agent can then use Wenlan's brief, recall, and capture tools.
01
Quick answer
Install Wenlan, then run ~/.wenlan/bin/wenlan connect cursor. Restart Cursor if needed, then verify that Wenlan tools appear and can recall a test capture.
Wenlan keeps durable context in its own local daemon while making it available to Cursor. That keeps the same source-backed context portable to Claude Code, Codex, and other clients later.
02
When this problem appears
Cursor doesn't carry memory between chats beyond Rules (its Memories feature was removed in 2.1), and Rules are Cursor-only. The gap appears when you want the same project context in Claude Code, Codex, Claude Desktop, or another MCP client.
03
Add Cursor as a client
Treat Cursor as an MCP client connected to one local Wenlan daemon.
- Install the Wenlan runtime once with the current setup path for your operating system.
- Run ~/.wenlan/bin/wenlan connect cursor.
- The generated Cursor config writes a global ~/.cursor/mcp.json entry.
- Restart Cursor if the MCP tools do not appear.
- Run ~/.wenlan/bin/wenlan doctor, confirm Wenlan is enabled in Cursor's MCP settings, and check Output → MCP Logs if tools don't appear.
- Ask Cursor Agent to use Wenlan context before a meaningful feature or bugfix, approving the MCP tool call if prompted.
- Capture why a design path was chosen, and build, CI, or setup gotchas once you have verified them.
- Recall by feature name, error string, or project decision.
- Capture decisions and handoff context, not raw chat history.
npx -y wenlan setup
~/.wenlan/bin/wenlan connect cursor
~/.wenlan/bin/wenlan doctor04
What to check next
If Cursor and another client both write memory, use spaces when contexts should stay separate. Cursor's agent can write pages itself; only background organization needs a separately configured model.
Try the local memory loop
Install Wenlan, connect your AI client, and verify that capture, recall, and handoff work on your machine.
FAQ