How it works

The memory server is a thin, stateless layer over the Memory API with its own token model.

#Transport and sessions

The server speaks Streamable HTTP. Each tool call is an independent API request; no conversation state is kept on the server.

#Identity resolution

An MCP token is bound to one identity. remember sets that identity as author and its default visibility; recall uses it as viewer. The model cannot change either.

#Tool calls map to API calls

Tool	API call
`remember`	`POST /v1/records`
`recall`	`POST /v1/recall`
`forget`	`DELETE /v1/records/{sourceId}`
`memory_status`	`GET /v1/jobs/{jobId}`

#Billing

Tool calls are billed as the API calls they make: ingestion credits for remember, and 0.1 credit per recall on Pay as you go.