MCP-Memory是一个Model Context Protocol服务器，为Claude Desktop、Cursor、Antigravity、Windsurf和Codex等AI代理提供持久化长期记忆能力[1]。该系统采用Open Knowledge Format（OKF v0.2）标准存储内存记录，并利用SQLite FTS5全文搜索引擎实现子20毫秒的键值查询和即时关键词搜索[1]。
该项目提供了四个主要MCP工具用于内存管理：保存内存、检索内存、搜索内存以及会话检查点管理[1]。内存记录自动落盘为markdown文件到memory/目录，SQLite数据库则存储在.mcp_memory/memories.db位置[1]。系统支持命名空间隔离、标签过滤、前置元数据验证和来源追踪等功能，可跨会话保存状态[1]。该服务通过自动化安装脚本与多个AI工具集成[1]。
MCP Memory, a Model Context Protocol server, has been introduced to provide AI agents with persistent long-term memory capabilities [1]. The system leverages the Open Knowledge Format standard (OKF v0.2) for storing memory records and employs SQLite FTS5 full-text search technology to enable rapid retrieval [1]. Supporting query response times under 20 milliseconds and instant keyword search functionality, MCP Memory addresses the need for agents to maintain and quickly access information across sessions [1].
The platform integrates with multiple AI development tools, including Claude Desktop, Cursor, Antigravity, Windsurf, and Codex [1]. It provides four core MCP tools: memory saving, memory retrieval, memory search, and session checkpoint management [1]. Memory records are automatically persisted as markdown files to a dedicated memory directory, while the underlying SQLite database is stored in a .mcp_memory/memories.db location [1]. The system includes namespace isolation, tag filtering, front-end metadata validation, and source tracking capabilities to organize and manage stored information effectively [1]. Installation and integration with supported AI tools are streamlined through automated setup scripts [1].