Memo
Give coding agents durable local memory with hybrid search—markdown on disk, sqlite-vec plus BM25—without sending recall to the cloud.
Overview
Memo is an MCP server for the Build phase that stores agent memory as local markdown with sqlite-vec and BM25 hybrid search, with no cloud backend.
What is this MCP server?
- Markdown-on-disk storage with sqlite-vec and BM25 hybrid retrieval
- MLX-native local stack—zero cloud dependency for memory operations
- PyPI package mlx-memo version 0.8.0 with stdio transport (runtime hint memo-mcp)
- Privacy-friendly recall for solo builders on Apple Silicon MLX paths
- Fits long-horizon agent projects that must remember decisions offline
- Package version 0.8.0 (mlx-memo on PyPI)
- Transport: stdio; registryType pypi
- Storage pattern: markdown-on-disk with sqlite-vec + BM25 hybrid
What problem does it solve?
Cloud-only memory forces you to ship sensitive project context off-device and breaks when you are offline.
Who is it for?
Solo builders on MLX-friendly setups who want private, offline-capable agent memory without a SaaS recall layer.
Skip if: Teams that need multi-user synced cloud memory or Windows-only workflows without Python/PyPI MCP setup.
What do I get? / Deliverables
Agents retrieve and update durable local memory through MCP using hybrid vector and keyword search on your disk.
- stdio MCP server exposing local read/write memory tools
- Hybrid sqlite-vec + BM25 search over agent-stored markdown
- Offline-capable memory layer with no required cloud API
Recommended MCP Servers
Journey fit
Local agent memory is built alongside your agent stack; you choose storage and retrieval during Build before scaling sessions in Grow or Operate. Agent-tooling subphase captures MLX-native MCP memory that agents query every turn, distinct from product analytics or uptime monitoring.
How it compares
Local hybrid-search memory MCP—not a hosted cross-tool context account like CacheTank.
Common Questions / FAQ
Who is Memo for?
Developers running local AI agents who need persistent markdown Memory and hybrid search without cloud recall services.
When should I use Memo?
Use it while building agent features when sessions must remember architecture decisions and notes across days, entirely on your machine.
How do I add Memo to my agent?
Install mlx-Memo from PyPI, configure the stdio MCP server with runtime hint Memo-mcp in your client, and ensure the on-disk markdown store path is writable.