Knowledge Graph Memory Server

The Memories With Lessons MCP Server is a lightweight, local knowledge‑graph backend that gives Claude (or any MCP‑compatible AI) the ability to retain user‑specific facts across sessions and learn from past mistakes through a structured lesson system. Traditional chatbot setups treat each conversation as an isolated event, making it difficult for the assistant to remember user preferences or to adapt its behavior after encountering recurring errors. This server solves that gap by persisting entities, relationships, observations, and lessons in a graph structure that the AI can query, update, and reason over at runtime.

At its core, the server exposes a set of CRUD‑style tools that operate on four primary concepts:

• Entities – Named nodes such as people, organizations or events. Each entity carries a type label and an arbitrary list of observations (atomic facts).
• Relations – Directed edges that connect two entities and describe their interaction in active‑voice terms (e.g., “works_at”).
• Observations – Single, self‑contained pieces of information attached to an entity. They can be added or removed independently, allowing fine‑grained updates.
• Lessons – Special entities that encode error patterns and their resolutions, complete with metadata (severity, environment, success rate) and verification steps. Lessons enable the assistant to remember why a particular solution worked and when it should be applied.

Developers benefit from this architecture in several concrete ways. In a customer‑support bot, the system can store user preferences (e.g., “prefers morning meetings”) and automatically surface them in future interactions. In a coding assistant, the lesson mechanism captures recurring build or runtime errors and teaches the AI how to fix them without manual re‑explanation. Because all data is stored locally, latency is minimal and privacy is preserved—no user data leaves the host machine.

Integration into an MCP workflow is straightforward: Claude issues a tool call such as or , the server updates the graph, and subsequent prompts can reference this enriched context. The active‑voice relationship syntax encourages natural language reasoning, while the lesson metadata allows the assistant to perform contextual checks (e.g., “Is this error happening on Windows?”) before suggesting a fix.

Unique advantages of the Memories With Lessons server include its lesson‑centric learning model, which turns every error into a reusable knowledge artifact, and its graph‑based persistence, enabling rich, interconnected memory that scales beyond flat key/value stores. For developers building AI assistants that need to remember context, adapt to mistakes, and operate securely on‑premises, this MCP server offers a powerful, low‑overhead solution.