amarodeabreu/claude-graph-memory ? reverse-engineered prompt

Build me a local memory layer for Claude Code that actually remembers my projects between sessions.

I want it to run fully on my machine, with Docker for the graph database, and connect to Claude Code through MCP so it can automatically read project docs and code structure, then store useful memories and decisions for later. It should index markdown docs, especially anything in a docs folder, and also understand basic code structure in Go, TypeScript, JavaScript, and Python. When I open a project, it should quietly build and update the graph in the background, so Claude can ask the right questions and find the right context without me repeating myself.

Please make it easy to install and remove with simple scripts, and include a way to verify that everything is working. I also want it to stay private, with nothing sent to the cloud, and to support multiple projects without mixing their data. If you need to check current docs for Claude Code or MCP details, look them up online.