MCP-Memory is a Model Context Protocol (MCP) server providing persistent long-term memory for AI agents like Claude Desktop, Cursor, and Codex. Memory records follow the Open Knowledge Format (OKF v0.2) standard and are indexed via local SQLite with FTS5 full-text search.

Key features include persistent state across sessions, dual-layer architecture with human-browseable OKF directories and high-performance SQLite indexing, namespace isolation, and zero-boilerplate setup via a Python wizard. The server exposes tools for storing, retrieving, searching, and managing memory records.

Storage uses project-isolated directories by default: OKF Markdown files in memory/ and SQLite database at .mcp_memory/memories.db. Environment variables allow customization of paths for global memory stores.

Setup involves cloning the repository, running the interactive setup wizard, and optional manual configuration via JSON or TOML files. Testing verifies OKF serialization and database operations. The system supports sub-20ms key lookups and instant keyword searches through automatic triggers and FTS5 indexing.