Summary
Memory you host, not memory a vendor holds.
Persistence lives in a Memory Bank API on your side, so what an assistant remembers about you sits in a database you can read and delete from. Two calls carry the weight: one writes, one asks.
What it is
A store-and-recall layer between an MCP client and your own Memory Bank API: write content with tags, search it semantically, and browse what has accumulated.
What you get
- Content stored with optional tags
- Semantic search over what has been stored, rather than exact-match lookup
- Stored memories browsed with filtering
Requirements
Node.js 18 or newer, run with npx memflow-mcp, and a Memory Bank API running for it to write to. MEMBANK_API_URL points at that server; an API key is optional. If the client reports spawn npx ENOENT, put the absolute path from `which npx` in the config.
Setup effort
One command — npx -y memflow-mcp