MCP / MEMORY
What MCP memory is
MCP memory usually describes a Model Context Protocol server that stores facts, preferences, or a knowledge graph so Claude Desktop, Cursor, or other clients remember you between chats. AIChatMemory is independent of those tools; its MCP is built for masked chat retrieval, not a personal notes graph.
Typical memory MCP servers
Many “memory” MCPs let an agent create entities, relations, and observations—often in local JSONL or SQLite. They compress habits and decisions into small records the model loads on the next turn. That helps when you want standing preferences, not a verbatim transcript from six months ago.
How that differs from a chat archive MCP
AIChatMemory uploads masked Claude Code, Codex, and Cursor transcripts from your Macs. The hosted MCP exposes three tools:
search— keyword plus meaning search across uploaded chats.count— exact totals for filters you care about.session— one masked chat returned in parts for the agent to read.
Agents retrieve; they write the answer themselves. There is no server-side “ask” tool. Chats are masked on your Mac before upload; reads of .env and credential files are dropped. Masking is not a guarantee that no sensitive text remains.
When to use which
Use a memory MCP when you want durable notes and rules. Use the archive MCP when you need proof from past coding-agent threads—deploy failures, refactors, or decisions buried in local JSONL on another Mac.
See also: Agent memory · Claude Code memory · All guides