An MCP memory server is a program that exposes save and recall memory operations to AI clients through the Model Context Protocol.
MCP (Model Context Protocol) lets an AI client such as Claude Desktop, ChatGPT Desktop, Codex or GitHub Copilot call tools on a separate server. A memory server implements tools like 'save this fact' and 'find memories about X'. Because the protocol is shared, the same server can serve several clients at once, and a memory written from one client is readable from another. That is the main appeal: one memory instead of one per app.
Memory servers differ in where they keep data. Cloud-hosted servers (Mnemoverse, Basic Memory Cloud) store it on the provider's side. Local servers store it on your machine and run as a process the client starts. Servers also differ in what else they offer: some are memory only; others add file search, document retrieval or knowledge graphs under the same connection. Setup is usually a JSON entry in the client's MCP configuration.
How Central Brain handles mcp memory server
memory_recall, memory_checkpoint, memory_compact) with read-only file tools (semantic_search, search_text, read_file, list_files). See the MCP memory server page.