Descriptions:
Fahd Mirza walks through the end-to-end setup of Memora, an open-source MCP (Model Context Protocol) server designed to give Claude Code and other AI agents persistent memory across sessions. The demo covers installing the package via pip, configuring the mcp.json file that Claude Code reads on startup, and connecting an optional knowledge graph visualization server on port 8765 — all demonstrated live on an Ubuntu server.
Beyond the installation walkthrough, Mirza offers substantive critical analysis of where persistent memory tooling currently falls short. He argues that Memora — like most similar tools — tries to do too much at once, layering in knowledge graph visualization and chat-with-memory features on top of what should be a narrow, well-executed persistence layer. He also flags a practical cost barrier: getting the tool running requires both a Claude Pro subscription at $20/month and a separate API key for an embedding model, making production deployment expensive at scale. A lighter option using a local TF-IDF embedding model is demonstrated but noted as a compromise.
For developers building agentic systems with Claude Code who have run into the session-amnesia problem — agents that forget project context, preferences, and prior decisions the moment a new session starts — this video provides both a working setup path and a realistic assessment of the current state of the art in open-source memory solutions.
📺 Source: Fahd Mirza · Published March 21, 2026
🏷️ Format: Hands On Build
