A personal MCP server that gives an AI assistant long-term memory — facts, preferences, past events, tasks, and notes — backed by Supabase (Postgres + pgvector) with hybrid semantic + full-text search.
A single Memory with:
- type —
fact,preference,memory,task,note - category — fixed 2-level vocabulary (e.g.
Work & Career > projects) - content — free text, embedded for semantic search
- optional due_date (tasks) and created_at override (backfill)
Strict per-user RLS. Searches require type or category_path to avoid semantic collisions.
- Supabase Edge Functions (Deno)
@modelcontextprotocol/sdkover Streamable HTTP (Hono)- Postgres + pgvector, OpenAI embeddings
add_memory, update_memory, delete_memory, search_memories, list_categories.