Memory MCP Server (Go)

Overview

The Memory MCP Server written in Go is a lightweight, high‑performance implementation of the Model Context Protocol that empowers large language models to maintain persistent knowledge graphs. By exposing a rich set of CRUD‑style tools for entities, relations, and observations, the server enables AI assistants to remember facts, track changes over time, and answer context‑aware queries across multiple sessions. This solves a core problem for conversational agents: the inability to retain structured memory without external databases or custom persistence layers.

At its heart, the server uses a SQLite backend with an automatic JSONL migration path. This design delivers low latency reads and writes while keeping disk usage minimal, making it suitable for both embedded devices and cloud deployments. The API supports creating, updating, and deleting entities with arbitrary types, defining relations in active voice (e.g., “Alice knows Bob”), and attaching time‑stamped observations that can be added or removed. A built‑in full‑text search powered by FTS5 offers quick node lookup, with a graceful fallback to simple string matching when necessary. These capabilities let developers model complex domains—such as customer histories, project dependencies, or knowledge bases—directly within the LLM’s context.

The server’s transport flexibility is another standout feature. It can operate over standard I/O, Server‑Sent Events with keep‑alive support, or streamable HTTP, allowing seamless integration into diverse AI workflows. Panic recovery in tool handlers ensures robustness, while optional sampling declarations let clients control text generation directly from the server. The pure‑Go SQLite implementation (no CGO) guarantees cross‑platform compatibility on Linux, macOS, and Windows.

Typical use cases include building a virtual assistant that remembers user preferences across days, maintaining an evolving ontology for a knowledge‑graph powered chatbot, or tracking changes in a collaborative document’s metadata. Developers can embed the server into existing MCP‑enabled assistants, letting the model query or mutate its memory with natural language prompts that are translated into structured tool calls. The result is an AI system that behaves as if it has a persistent, queryable brain—significantly enhancing continuity and personalization in conversational experiences.