Does Cursor have memory? What happened to Cursor Memories

Does Cursor have memory?

So the agent in your editor learns nothing on its own between chats any more. It knows what a rule says. The two places where Cursor agents still write their own notes are Automations and Projects, and both run on Cursor's Cloud Agents.

What were Cursor Memories?

Facts Cursor saved from your chats and fed back into later ones. The 1.0 release (4 June 2025) introduced them as a beta: “Cursor can remember facts from conversations and reference them in the future,” stored “per project on an individual level” and switched on under Settings, Rules. The 1.2 release (3 July 2025) made them generally available and added “user approvals for background-generated memories”: a memory Cursor drafted on its own waited for you to accept it. Users found them under Settings, Rules & Memories, Saved memories (forum, July 2025).

Why are my Cursor memories gone?

Because the feature is. A Cursor staff reply on the forum on 25 November 2025 says “The Memories feature was intentionally removed starting from version 2.1.x”. The 2.1 release notes (21 November 2025) do not mention it, and no reason has been published. The old docs page, docs.cursor.com/context/memories, now redirects to the docs home. If a guide tells you to switch on memory generation, or to fix “Cursor memories not working”, it describes Cursor 1.x.

How do I export Cursor memories to rules?

The same reply gives the steps. Press Cmd+Shift+P, run Export memories, and Cursor saves them to an .mdc file. Paste what you keep into a User Rule or a Project Rule. Downgrading to 2.0.77 brought them back too, but the reply warns that works only “before older versions lose support.” Those steps are from November 2025: if your build no longer has the command, the memories are not coming back through the editor.

Read the export before you paste it. A memory was one fact from one chat, and a rule applies to every chat after it. Our Cursor rules best practices cover which apply mode each fact should get.

How do I use memory in Cursor now?

Write it down where Cursor reads it. From the rules docs:

• Project rules are .mdc files in .cursor/rules, under version control, each set to Always Apply, Apply Intelligently, Apply to Specific Files or Apply Manually. A plain .md file there is ignored. Type /create-rule in Agent and describe the rule, which is the closest thing left to “remember this”.
• User rules live in Customize, then Rules, and apply to all your projects in Agent. Rules do not affect Cursor Tab.
• AGENTS.md works in the project root and in subfolders. A nested file applies to its folder, and the more specific one wins.
• Team rules are set in the dashboard on Team and Enterprise plans. An admin can make one required, and then nobody can turn it off.

Cursor asks you to keep each rule under 500 lines and to split big ones into smaller rules.

What are Cursor Automations memories?

Notes an automation keeps between its own runs. Automations launched on 5 March 2026 with “a memory tool that lets them learn from past runs”. The Automations docs say the agent can “read and write persistent notes across runs for the same automation”, saved as MEMORIES.md by default and stored outside the agent's working files. They are on by default. You can view, edit and delete them in the tool configuration, and the agent can delete stale ones itself.

Two limits follow from the docs. They are scoped to one automation, so another automation, or a chat in your editor, starts without them. And Cursor warns that an automation reading untrusted input “may lead to misleading or malicious memories” in later runs, the risk our note on agent memory poisoning covers.

What is shared context in Cursor Projects?

The nearest thing to the old Memories. Projects went into beta on 10 September 2026. “Each Project maintains a set of files that sync across every cloud and local machine its agents use,” and “agents add research and artifacts, along with what they learn about the codebase and how you prefer work to be done.” The difference is the owner: a Memory belonged to you in one project, while shared context belongs to the Project.

The Projects docs list it for Premium and higher, Teams, Enterprise and legacy paid plans, and Enterprise needs Cursor 3.21.9 or later. It is not on Free, Plus or the legacy Hobby plan, and not with Privacy Mode (Legacy), because Projects run on Cloud Agents.

Does Cursor have memory like Claude Code?

Not in the editor. Claude Code has auto memory: notes Claude writes for itself, on by default, kept in a folder on your machine (our Claude Code auto memory note has where and how to turn it off). Since 2.1, a local Cursor agent writes a memory only when you ask it for a rule. Codex has memories too, off by default (Codex memories), and so does GitHub Copilot (Copilot Memory). The side-by-side comparison puts all four in one table.

What is a Cursor memory bank?

A habit, not a Cursor feature. It comes from Cline's Memory Bank: a memory-bank/ folder of six markdown files (projectbrief.md, productContext.md, activeContext.md, systemPatterns.md, techContext.md, progress.md) and an instruction to read all of them at the start of every task. In Cursor you commit the folder and add an Always Apply rule that says to read the files and update them. It works, it costs every file in every task, and it stays true only as long as someone keeps the files current.

Can I add memory to Cursor with MCP?

Yes. A memory server is an MCP server like any other: put it in .cursor/mcp.json for one project or ~/.cursor/mcp.json for all of them (MCP docs). The agent then has tools to save and search, and it uses them when it decides to, so a rule that says when to search is usually part of the setup. Our list of MCP memory servers compares the options.

Does Cursor memory reach Claude Code, Codex or Copilot?

No means the tool's own instruction docs do not list it. Codex reads AGENTS.md and AGENTS.override.md. Copilot reads its own instruction files, AGENTS.md, and a root CLAUDE.md or GEMINI.md. Claude Code reads AGENTS.md since v2.1.277, by default only in a project with no CLAUDE.md (details). So a rule your whole team needs in every tool belongs in AGENTS.md, not in .cursor/rules.

How teams share what Cursor learns

Every option above holds what somebody wrote down, or what one tool's agents wrote for that tool. What a teammate decided in Slack or in a review never reaches a rule unless someone copies it there. Harbor is one way to close that: shared memory for AI coding agents. It learns from where your team decides (Slack, pull request reviews, docs and agent sessions), lets you set what waits for a person, and gives Claude Code, Cursor, Codex and GitHub Copilot the rules that apply to each task. To use the same rules in every agent without it, see how to share memory and rules across tools.