MCP-Memory provides AI agents with persistent memory using OKF and SQLite FTS5

Persistent Memory for AI Agents

MCP-Memory is a Model Context Protocol (MCP) server designed to equip AI agents, including tools like Claude Desktop, Cursor, and Codex, with persistent, long-term memory. This allows agents to store, retrieve, and manage stateful memory snippets that remain available across multiple chat turns and sessions, addressing a common limitation in current AI agent architectures.

Open Knowledge Format and SQLite Indexing

Memory records within MCP-Memory are formatted according to the Open Knowledge Format (OKF v0.2) standard. Each memory item is stored as a Markdown document with YAML frontmatter, adhering to specified OKF rules. For efficient data retrieval, these records are indexed using a local SQLite instance, which leverages FTS5 (Full-Text Search) for rapid key-value lookups and content searches, achieving sub-20ms lookup times.

Dual-Layer Architecture and Features

The system employs a dual-layer architecture. It automatically syncs every memory to disk as a raw .md file within a human-browseable OKF directory, complete with hierarchical index.md files and log.md for update history. Key features include namespace isolation for contextual separation (e.g., user preferences, project architecture), and a setup wizard for auto-configuration with installed MCP tools. The server exposes primary MCP tools for agents to store, update, and query memory records using parameters like key, content, tags, and namespace.