Shared Cursor memory: one team memory for every agent
Last updated September 25, 2026.
Cursor forgets. Every thread starts over.
Whatever your Cursor agent learned in the last thread (conventions, architecture, decisions, the gotcha that cost an afternoon) doesn't exist in the next one. Multiply that by a team: five developers, five agents, each relearning the same codebase separately, none of them benefiting from what the others' agents already figured out.
Rules files help, but they're flat, hand-maintained, and per-repo. What a team actually needs is memory that accumulates on its own and is shared by default.
What Cursor team memory looks like
Memento (mementoagi.com) is hosted memory your agents connect to over MCP (https://api.mementoagi.com/mcp, with an API key; one mcp.json entry, since Memento is not in the official Cursor marketplace yet).
With a team org:
- Every teammate's agent reads and writes one shared memory of the codebase. When one person's agent learns why the billing service is structured the way it is, everyone's agent knows it in their next session.
- A new developer's agent starts day one already knowing the architecture, the conventions, and the gotchas, because the team memory already holds them.
- The memory is plain markdown pages, organized hierarchically, that humans can read and edit on the web. If the AI has something wrong, anyone on the team can open the page and fix it, and the fix propagates to every agent.
- It works across tools: the same shared memory serves Cursor, Claude Code, Windsurf, and Copilot, and follows people across machines.
You create an org, invite teammates by email, and drop shared knowledge into the org scope. Personal memory stays personal by default.
Other hosted options for team memory
Memento isn't the only way to do this, and depending on your shape one of these may fit better: Supermemory (hosted memory API with MCP), Context Cloud (shared workspaces with RBAC), Zep (governed enterprise context), and Mem0's hosted tier (memory API for building your own stack). Claude-Mem is local by default with a hosted CMEM Cloud plan. The comparison page has the full table with dated facts.
Memento's difference is legibility: the shared memory is documents your team can read, not embeddings behind an API. Your agents and your humans see the same pages.
Set it up
- Create a free account at mementoagi.com and get an API key.
- Add the Memento MCP server to Cursor with one mcp.json entry (the Getting Started guide has it ready to paste).
- Create an org, invite your team by email, and start working. The memory builds itself from there.
Free tier available: https://mementoagi.com