mcp-claude-memory

Smart, self-maintaining, cross-machine memory for Claude Code — Postgres-backed, no rented LLM brain.

mem0 and Honcho are memory stores that pay a separate LLM (OpenAI) to do the thinking. mcp-claude-memory flips that: in Claude Code the LLM is already in the room, so Claude does the extraction, reasoning, and background reconciliation itself — for free. The backend is just Postgres.

npm install -g mcp-claude-memory
mcp-claude-memory init

What you get

• Follows you across machines — every machine points at one shared Postgres, so office / home / laptop share a single brain. Postgres is the sync.

• Works in every project — registered at user scope, so the memory tools are available everywhere, not per-repo.

• Automatic or manual capture — choose auto (Claude saves and updates memories silently as you work, no commands) or manual (only when you ask). Manual remember/forget work in both modes.

• Semantic recall — optional embeddings (OpenAI or local Ollama, no key required) give vector search blended with full-text via Reciprocal Rank Fusion. Without embeddings it falls back to Postgres full-text — it works with just a Postgres URL.

• Self-healing (living memory) — every fact tracks source, confidence, and freshness. Confidence grows as facts are re-confirmed; the review tool surfaces stale memories; contradictions get superseded.

• Background dreaming — mcp-claude-memory dream archives duplicates and runs a headless Claude pass to reconcile stale/contradictory memories. Schedule it with cron.

• Yours — plain rows in your own Postgres, no opaque vendor store, no required LLM API key.

Related MCP server: claude-crowed

Setup

npm install -g mcp-claude-memory
mcp-claude-memory init     # global setup, then restart Claude Code

The wizard asks for:

• Postgres connection string (Supabase / Neon / Railway / self-hosted all work).

• Namespace — your identity; the same namespace on another machine shares one brain.

• Capture mode — auto (recommended) or manual.

• Embeddings provider — none (full-text only), openai, or ollama.

It migrates the schema, registers the MCP server at user scope, and (in auto mode) installs a small instruction block into ~/.claude/CLAUDE.md.

Add another machine by running the same init with the same Postgres URL, namespace, and capture mode.

Note: Switching embedding provider or model does not re-embed existing memories — older memories remain searchable via full-text until they are saved again.

Memory tools (used by Claude)

In auto mode Claude calls these proactively; you can also just say "remember this" / "forget that".

Dreaming (background maintenance)

mcp-claude-memory dream                # run a maintenance pass now
mcp-claude-memory dream --print-cron   # print a crontab line to schedule it

The pass archives exact-duplicate memories, then launches a headless Claude pass (claude -p) that reviews and reconciles stale/contradictory memories. The semantic pass needs the claude CLI logged in on the machine running the job; if it's unavailable the duplicate cleanup still runs and the semantic step is logged and skipped.

Requirements

• Node ≥ 20.

• A Postgres database. For semantic embeddings, Postgres with the pgvector extension (Supabase / Neon provide it; self-hosted needs CREATE EXTENSION vector, done automatically when you enable a provider).

License

MIT