Persistent semantic memory MCP server for Claude Code — powered by Supabase + pgvector + Ollama.
Replaces keyword search with real semantic similarity so you can query with natural language across sessions.
claude plugin marketplace add rnblanco/supa-brain && claude plugin install supa-brainFull setup guide (Ollama, Supabase, credentials): plugin/claude-code/README.md
| Tool | Description |
|---|---|
mem_save |
Save an observation with semantic embedding |
mem_search |
Search by natural language query |
mem_session_summary |
Summarize and persist a session |
mem_get_observation |
Retrieve observation by ID |
mem_update |
Patch an existing observation by ID |
mem_delete |
Delete observation by project + topic_key |
mem_context |
Get recent sessions and observations (no embedding cost) |
mem_suggest_topic_key |
Generate a stable slug for upserts |
mem_capture_passive |
Extract learnings from Key Learnings sections |
Create ~/.supa-brain/config.env:
# Supabase Dashboard → Settings → Database → Connection string (URI mode)
DB_URL=postgresql://postgres.YOUR_REF:YOUR_PASSWORD@aws-0-REGION.pooler.supabase.com:6543/postgres
# Optional — only needed for the db:migrate command
SUPABASE_URL=https://YOUR_REF.supabase.co- Ollama running locally with
nomic-embed-textmodel - Supabase project with pgvector extension enabled
- Claude Code >= 1.5.0