brayanfz013/memory-graph-mcp — reverse-engineered prompt

Build me a local MCP memory server for AI coding tools like Claude Code, Cursor, and Codex. I want it to keep a per project memory on disk so agents can remember past decisions between sessions, search that memory semantically, store findings as typed graph nodes with relationships, and automatically turn important items into simple wiki style pages. It should also support shared short lived state and caching so multiple agents can coordinate without repeating expensive work.

Please make it workspace scoped, using a local database inside the repo, with local embeddings by default so it works without cloud keys. It should expose a small set of useful tools like recalling past work, recording a finding after solving something, fetching a wiki page, exploring graph links, and checking memory health or stats. Make it easy to install as a Python package and wire into any MCP client, and also give it a smooth Claude Code plugin setup with a helpful slash command, examples, and basic troubleshooting. Look up current MCP docs online if you need to.