Memory MCP Server
by Model Context Protocol (reference)
Reference MCP server for persistent agent memory using a local knowledge graph. The starter every memory-aware agent reaches for.
MCP
Memory MCP Server
Added 17 May 2026
Overview
Memory MCP is a reference implementation in the modelcontextprotocol/servers repo. It stores entities, relations, and observations in a local graph so agents can remember across runs. Tiny, opinionated, easy to extend into a real long-term memory layer.
Best for
Best for
Anyone who wants 'the agent remembers me' without a heavy memory stack
Use cases
- Persist user facts across Claude or Cursor sessions
- Build an assistant that actually remembers project context
- Bootstrap an agent memory layer without picking a vector DB on day one
- Template for your own memory schema
Notes
Why it matters
Persistent memory is the difference between an agent that helps and an assistant that knows you. The reference MCP is the cheapest way to feel that shift.
How teams use it in production
Start with the reference. Fork into a server backed by Postgres or Supabase once you have enough writes to outgrow the local store.
What to watch
The convergence between MCP memory servers, vector stores, and per-user graphs is the real long-term memory architecture question for agents.
Pros
- Reference quality, kept current
- Graph model is more queryable than free-text notes
- Tiny footprint, easy to inspect
- Forks cleanly into product-specific memory schemas
Cons
- Local storage means no cross-device by default
- No vector retrieval out of the box
- Graph editing UX still rough
Pairs with
Other entries in the index that connect to this one. Click through to see the chain.
Claude Code
Anthropic
Anthropic's terminal-native coding agent. Reads your repo, edits files, runs tests, ships PRs.
Cursor
Anysphere
The AI-first code editor. Tab to autocomplete, Composer to multi-file refactor, Agents for the long-running stuff.