Memory Vault
Give your agent durable, local-first memory with hybrid vector search backed by Postgres and pgvector you control.
Overview
Memory Vault is an MCP server for the Build phase that provides a local-first AI memory layer with hybrid search on self-hosted Postgres and pgvector.
What is this MCP server?
- Local-first AI memory with hybrid search on Postgres + pgvector
- Self-hosted MIT stack; OCI image ghcr.io/mihaibuilds/memory-vault-mcp
- Stdio MCP transport with required DB_HOST and configurable DB_PORT, DB_NAME, DB_USER
- Docker-friendly host.docker.internal or compose service naming
- Version 1.0.6 published on OCI registry
- MIT license (per description)
- OCI package ghcr.io/mihaibuilds/memory-vault-mcp:1.0.6
Community signal: 4 GitHub stars.
What problem does it solve?
Agents forget project context between sessions unless you paste the same background every time or trust opaque cloud memory.
Who is it for?
Privacy-minded solo builders who already run or can run Postgres+pgvector and want agent memory under their control.
Skip if: Builders who want zero-ops hosted memory with no database administration.
What do I get? / Deliverables
Your agent reads and writes durable memories in your Postgres instance through MCP with hybrid semantic retrieval.
- MCP-backed read/write memory store for agents
- Hybrid search over embedded memories in your database
- Repeatable self-hosted memory layer across projects
Recommended MCP Servers
Journey fit
Build integrations is the canonical shelf because you stand up Postgres, env vars, and the MCP bridge; memory then supports every later phase. Integrations fits persistent memory as infrastructure your agent calls—not a one-off frontend task.
How it compares
Self-hosted memory database MCP, not a hosted RAG SaaS or a single chat export skill.
Common Questions / FAQ
Who is Memory Vault for?
Solo builders and small teams running AI coding agents who need persistent, self-hosted memory with vector search.
When should I use Memory Vault?
Use it when cross-session context, preferences, and project facts should live in your own Postgres rather than ephemeral chats.
How do I add Memory Vault to my agent?
Deploy Postgres with pgvector, set DB_HOST, DB_PORT, DB_NAME, and DB_USER, then register the OCI image ghcr.io/mihaibuilds/memory-vault-mcp:1.0.6 with stdio in your MCP config.