Cuba Memorys
Give coding agents durable project memory in PostgreSQL with ranked retrieval, sync folders, and optional judge-backed consolidation.
Overview
cuba-memorys is an MCP server for the Build phase that provides 25 tools for PostgreSQL-backed persistent agent memory with BM25+MMR+OOD retrieval and audit trails.
What is this MCP server?
- 25 MCP tools for store, search, sync, and governance of long-lived agent memory
- Retrieval stack combines BM25, MMR, and OOD-style ranking for less repetitive context
- PostgreSQL-backed knowledge graph with CFR-21-oriented audit hooks for traceability
- stdio npm package with DATABASE_URL and optional cuba_juez judge backends (CLI, API, heuristic, auto)
- Project scoping via CUBA_PROJECT_FILTER and file sync via CUBA_SYNC_DIR (default ./.cuba-memorys/)
- 25 MCP tools documented for memory operations
- Registry version 0.9.3; npm identifier cuba-memorys
- Requires DATABASE_URL; default sync dir ./.cuba-memorys/
Community signal: 24 GitHub stars.
What problem does it solve?
Agents forget repo context between sessions and flood prompts with duplicate chunks when you only use the chat window.
Who is it for?
Solo builders running Claude Code or Cursor daily who want self-hosted, tool-driven memory with PostgreSQL control.
Skip if: Quick prototypes with no database, or teams that cannot run Postgres and maintain an npm MCP server.
What do I get? / Deliverables
Memory lives in a queryable store your agent can search, scope by project, sync to disk, and audit instead of re-explaining every task.
- Running stdio MCP memory service with 25 agent-facing tools
- Project-scoped memory store with optional .cuba-memorys sync exports
Recommended MCP Servers
Journey fit
How it compares
Self-hosted memory MCP with graph and retrieval—not a cloud-only notes plugin or a single SKILL.md prompt trick.
Common Questions / FAQ
Who is cuba-memorys for?
It is for agent-first developers who need durable, searchable memory in PostgreSQL with MCP tools rather than ad-hoc markdown dumps.
When should I use cuba-memorys?
Use it while building agent workflows and across later operate iterations whenever sessions must recall decisions, entities, and project-scoped facts.
How do I add cuba-memorys to my agent?
Install the npm package cuba-memorys, set DATABASE_URL (and optional CUBA_* env vars), and add a stdio MCP server entry in Claude Code, Cursor, or Windsurf.