search and add are the whole surface; what makes it usable at scale is that a memory carries tags, the people it involves and a topic, and search can filter on all three plus a time window. Because it is a hosted endpoint with OAuth, the same memory follows you between Claude Desktop, VS Code and anything else that speaks MCP.
A hosted memory service reached over MCP. Memories are stored with metadata — tags, people mentioned, a topic category — and searched semantically with those fields as filters. The same store is reachable from the Python and TypeScript SDKs and from the web app.
- search — semantic search over your memories, narrowed by tags, people_mentioned, topic_category or a temporal_filter such as today or this_week
- Threshold and retrieval_limit on search, so you can trade recall against precision
- add — store a memory with tags, people mentioned, a topic category and any extra metadata you want to keep
- Project isolation: the OAuth token scopes every call to the project you selected, so work and personal memories stay apart
A SelfMemory account. The server is remote at https://mcp.selfmemory.com/mcp over Streamable HTTP, and authorization runs through OAuth 2.1 — your client discovers the authorization server from the protected-resource metadata and walks you through login in the browser. The scopes are memories:read and memories:write. Nothing to install locally.
One command — pip install selfmemory