DEV Community

Cover image for How to give Cursor memory between sessions
nestdaddy
nestdaddy

Posted on Originally published at nestdaddy.com

How to give Cursor memory between sessions

You spent the afternoon with Cursor's agent on a refactor. You agreed on the folder layout, picked a library and stopped halfway through the migration. Next morning you open a new chat and type "continue the migration". The agent starts reading the codebase from scratch, and suggests the library you rejected yesterday.

Cursor is brilliant inside one session. Between sessions it forgets.

What Cursor remembers today

  • Rules files. Project rules (in .cursor/rules) and user rules are added to the agent's context. They are great for stable things: code style, the stack, "always write tests".
  • The codebase itself. The agent can search and read your files every time.

What neither of them holds: decisions and why you made them, what you did yesterday, what you tried that failed, and anything from your other AI apps. Rules only contain what you type by hand, and nobody updates a rules file after every session.

A memory the agent writes itself

NestDaddy Memory is an MCP server. Once it is connected, Cursor's agent can:

  • save decisions as you make them ("we use Postgres, because orders and stock need one transaction"),
  • find them by meaning in any later session, even when you word things differently,
  • replace a decision when you change your mind, so the old plan never comes back,
  • look back in time: "what did we change last week?" gets a day-by-day answer,
  • find past conversations: "continue yesterday's work" finds where you stopped.

And because it is one memory for every AI you use, what you decided in Claude or ChatGPT this morning is there in Cursor this afternoon.

Connect it to Cursor

  1. Start your free month at nestdaddy.com/memory. No card.
  2. In Cursor, open Settings and find MCP (under Tools in recent versions).
  3. Add a new MCP server. Choose the remote or HTTP type, name it nestdaddy-memory and use this address: https://nestdaddy.com/memory/mcp
  4. Your browser opens a NestDaddy page. Sign in and click Allow. No keys to copy.

The same steps work in VS Code, Windsurf and other editors that support remote MCP servers.

Make the agent use it every session

Add one line to your user rules (or a project rule), so the agent checks memory before it starts:

My main memory is the NestDaddy Memory MCP server. On every task, first call recall with my request, and search it before you decide anything about this project. When we decide something, save it with save_memory. After each answer, log the turn with log_turn.

Now try it. Tell the agent: "Remember that we chose Postgres for this project, because orders and stock need one transaction." Open a new chat tomorrow and ask: "Which database did we pick, and why?"

Why not just a markdown file?

A memory.md in the repo is a fine start. It breaks down when:

  • it grows past what fits in context, so the agent stops reading it all,
  • two decisions disagree and the agent cannot tell which one is newer,
  • you want it in Claude, ChatGPT or on your phone, not only in this repo.

NestDaddy Memory handles those for you: search by meaning instead of reading everything, the latest decision always wins with full history kept, and one memory across every AI and device.

Start today

Every session you work without memory is context you will explain again. Start your AI's memory today, and let your agent pick up exactly where you left off: nestdaddy.com/memory.


Originally published on the NestDaddy blog. NestDaddy gives your AI today's internet: live web search and real-time breaking news from its own index. nestdaddy.com/mcp

Top comments (0)