MCP Server for a remote Memory Bank

Showcase
1

I have been experimenting a solution I made on top of Cline’s concept of what they call memory bank:

I have been managing to use it with cline & cursor (I use both in Cursor.)
After that I made an extension for me to be able to watch md files in the source of the memory bank.

I have the custom instructions on both cline & cursor. It has been working pretty well.

I create some custom prompts to do specific tasks. The tasks rules are refined within the bank, any agent contributes to improve the bank.

The results have been impressive!

Next step will be to have hybrid RAG + option to connect to a data-base locally or remotely (for teams to share the same bank)

Another advantage of a remote memory bank is that when you work on a distributed system you can take lots of advantage in having agents consulting rules of other layers of the system to implement tasks.

For specific tasks create custom files.

Tell me what you think. Try it out!

I will publish it as an npm package

6 Likes
2

Does your MCP Memory Bank just save the correspondence? Or is it able to create the project description, logic structure, name of classes, methods, functions and variables by itself?

3

the agent does it by itself. it’s within the custom instructions. take a look. But ofc, let’s say you open a huge project, then the agent might not create the core files based on the overall context, as it would be ridiculous to ask the agent to look up a whole repo to write the files. But, as you progressively work on a project, these files get better and better, as the agents learn more about it and this is the proper way to go imo

4

Couldn’t figure out how to use your MCP (
How to verify that the MCP is working.
Where to add cline rules.
Is it possible to control the language in which the documentation will be generated?