Memory MCP Server
Persist decisions, entities, and relationships across agent sessions with SQLite-backed memory, search, and graph traversal.
Overview
Memory MCP Server is a MCP server for the Build phase that provides SQLite-backed persistent memory with full-text search and graph traversal for agents.
What is this MCP server?
- SQLite-backed persistent memory with full-text retrieval
- Graph traversal for linked entities and relationships
- npm package @j0hanz/memory-mcp v1.7.0 with stdio MCP transport
- Survives session restarts unlike ephemeral chat context alone
- Fits solo-builder stacks that track product decisions, customer notes, or architecture facts
- Version 1.7.0, npm package @j0hanz/memory-mcp
- Storage: SQLite with full-text retrieval and graph traversal
- Transport: stdio per MCP server schema 2025-12-11
What problem does it solve?
Agents forget prior decisions and entity relationships every session, forcing you to re-explain architecture and customer context.
Who is it for?
Indie builders running multi-week agent projects who need durable, queryable memory beyond the current chat thread.
Skip if: One-off scripts or teams that already centralize knowledge in a audited team wiki with strict compliance requirements only.
What do I get? / Deliverables
After install, agents can store, search, and traverse linked memories so long-running solo projects keep coherent context.
- Persistent memory records searchable via full-text MCP tools
- Linked entities traversable as a graph from agent sessions
- Reduced re-onboarding of project context after restarts
Recommended MCP Servers
Journey fit
Long-lived agent memory is introduced during Build agent-tooling setup but pays off across Operate and Grow when context must survive restarts. Agent-tooling is the first shelf where builders wire durable memory MCP alongside filesystem and fetch tools.
How it compares
Persistent memory MCP with FTS and graphs, not a vector SaaS RAG platform or a simple notes markdown skill.
Common Questions / FAQ
Who is memory-mcp for?
Solo developers and agent-first workflows that need cross-session recall of decisions, entities, and relationships via MCP.
When should I use memory-mcp?
Use it when the same project spans many sessions and you want the agent to remember facts, link concepts, and full-text search prior notes.
How do I add memory-mcp to my agent?
Install @j0hanz/memory-mcp from npm, register the stdio server in your MCP client, and point it at a persistent SQLite storage location per the repo README.