I think that a great feature of Chat or Composer could be marking a certain corrective statement from a user as a memory to store or add to the project’s .cursorrules file. Some enthusiasts are currently trying to implement this using some hacks and workarounds. But e.g. ChatGPT has memory. Can we have it in Cursor, in a model-independent way? Or expose access to such functionality at a model level for models that support it?
this is actually a great idea, imagine cursor learning about your codebase and coding style as you use it, instead of having to create and maintain a bunch of md files and rules. it’d just naturally pick up on how you do things
I’m using pinecone MCP for that: GitHub - sirmews/mcp-pinecone: Model Context Protocol server to allow for reading and writing from Pinecone. Rudimentary RAG
Works quite well for me. You can ask the agent to save to the memory and then retrieve when needed. The only blocker is a yolo mode for MCP tools.
till they implement this in cursor, i believe this is the best workaround rn:
Cline has something like this, but it seems more “natural”
I made a cross client compatible solution: @allpepper/memory-bank-mcp - npm
You can have multiple clients/agents to increment the bank, like using cline, cursor, w1ndsf…, claude, aider + mcp and share this bank for everybody.
I am now focusing in researching for a remote solution to store the bank so it can be cross team compatible.
Lot’s of options, though… mongodb + native vectorial store, multi db support, etc.
The extension simply saves the history of communication with the AI. It also allows you to unshare them. It does not form a specific project description.
Cursor has removed the addition of MCPs via the GUI. Please add an example of adding via .cursor\mcp.json.