Knowledge Graph Memory Server

Overview

The Mcp Memory Py server provides a lightweight, persistent knowledge‑graph backend that lets Claude and other Model Context Protocol clients remember user‑specific facts across sessions. By exposing a set of CRUD‑style tools that manipulate entities, relations, and observations, it turns transient chat context into a durable memory store. This solves the common problem of stateless AI assistants that forget prior conversations, enabling richer, more personalized interactions.

The server is built around three core concepts: entities, relations, and observations. Entities are named nodes (e.g., John_Smith or Anthropic) that carry a type label and an arbitrary list of observations. Relations are directed edges in active voice (e.g., works_at) that connect two entities, capturing how they interact. Observations are atomic strings attached to a single entity, representing discrete facts that can be added or removed independently. Together these structures form a graph that can be queried, updated, and traversed through the MCP tool set.

Key capabilities include:

• Entity and relation lifecycle management (create, delete, bulk operations) with built‑in deduplication.
• Fine‑grained observation handling that allows adding or removing single facts without affecting other data.
• Graph retrieval and search tools (, , ) that let an assistant fetch the entire memory, locate specific nodes by keyword, or pull a subset of related entities.
• Atomic operations that preserve data integrity; for example, attempting to add an observation to a non‑existent entity fails cleanly.

In real‑world scenarios, developers can use this server to build context‑aware assistants that remember user preferences, project details, or domain knowledge across conversations. For example, a virtual research assistant could retain information about a user’s publication history, collaborators, and ongoing projects, enabling it to suggest relevant papers or schedule meetings without repeated prompts. Likewise, a customer support bot could persist user ticket history and product usage patterns to deliver faster, more tailored assistance.

Integration into AI workflows is straightforward: an MCP‑enabled client calls the appropriate tool to persist new facts, then later queries or updates the graph as needed. The server’s stateless HTTP interface means it can run alongside other MCP services, scale independently, and be swapped out for alternative backends (e.g., a graph database) without changing client code. Its Python implementation also offers easy extensibility for custom logic or persistence layers.

Overall, Mcp Memory Py stands out by combining a simple, graph‑based memory model with robust MCP tooling, giving developers an efficient way to add long‑term context to conversational AI applications.