xgmem MCP Memory Server

Overview

The xgmem MCP Memory Server is a TypeScript‑based service that extends the Model Context Protocol by providing a structured, persistent memory layer for AI assistants such as Claude. Instead of treating knowledge as transient prompt text, xgmem organizes information into a lightweight knowledge graph consisting of entities, relations, and observations. This design allows agents to store context in a way that mirrors real‑world relationships, enabling more natural reasoning and recall across multiple projects.

Solving the Memory Problem

Large language models excel at generating text, but they lack an internal state that survives beyond a single conversation. Developers often need to persist domain knowledge—facts about people, products, or processes—to maintain continuity and avoid re‑learning. xgmem addresses this gap by offering a dedicated memory store that can be queried and updated via MCP tools. By keeping data on disk in files, it guarantees durability and cross‑session availability without requiring external databases or complex infrastructure.

Core Capabilities

• Knowledge Graph Storage: Persist entities, relations, and observations in a graph format that supports hierarchical and relational queries.
• CRUD via MCP Tools: Expose intuitive tools (, , , etc.) that map directly to graph operations, making it simple for agents to write or read memory.
• Project Isolation & Sharing: Each project has its own namespace, yet the server can copy or migrate memory between projects with tools like , facilitating knowledge reuse.
• Disk Persistence: All data is serialized to the directory, ensuring that memory survives restarts and can be version‑controlled if desired.
• Docker & TypeScript Friendly: The server ships with Docker support and a TypeScript codebase, allowing developers to integrate it into CI/CD pipelines or local development environments effortlessly.

Real‑World Use Cases

• Agent Ecosystems: Multiple LLM agents working on the same domain can share a unified memory, reducing duplication and improving consistency.
• Project Knowledge Transfer: When onboarding new teams or migrating projects, the tool can duplicate relevant knowledge graphs, speeding up ramp‑up time.
• Audit & Traceability: Observations can be logged with timestamps and sources, providing a clear audit trail for compliance‑heavy industries.
• Personal Assistants: A personal AI can maintain a persistent graph of user preferences, appointments, and past interactions, delivering more personalized responses over time.

Integration into AI Workflows

Developers configure xgmem in their MCP configuration, then invoke its tools from any MCP‑compatible client. An agent can, for example, call after completing a task to record new facts, or query to retrieve context before generating a response. Because the server exposes standard MCP tools, it plugs seamlessly into existing pipelines without requiring custom adapters or SDKs. The ability to read the graph () also lets developers visualize memory, aiding debugging and model tuning.

Distinct Advantages

xgmem’s graph‑centric approach stands out among memory servers that rely on flat key/value stores. By modeling entities and their relationships, it enables richer inference patterns—agents can traverse connections to discover related facts or detect inconsistencies. The server’s lightweight, Docker‑ready design lowers the barrier to entry, making it suitable for both prototyping and production deployments. Finally, its explicit support for project namespaces combined with memory copying tools gives developers fine‑grained control over knowledge ownership and sharing, a feature rarely found in standard MCP memory services.