Engram
A local-first memory MCP server that enables storing, searching, and managing personal memories with hybrid keyword and semantic recall, all on-device.
README
Engram
Your memory, on your machine.
Every AI platform is racing to build "memory" — and locking it inside their own walls. Your context, preferences, and history end up trapped in one vendor's product, invisible and unportable. Engram is the opposite bet: an open, local-first memory layer that you own, that follows you across every tool.
Engram runs as a Model Context Protocol server, so any MCP-capable client (Claude Code, Claude Desktop, and more as they adopt the protocol) can read and write your memory. Everything stays on your machine — the store, the search, even the embedding model. Nothing leaves.
How it works
- Local-first — a single SQLite file on your disk is the source of truth.
- On-device embeddings — semantic search runs locally (bge-small); your memories are never sent to a third party.
- Hybrid recall — keyword (FTS5) fused with semantic similarity, so it finds what's relevant, not just what shares words.
- Stays coherent — flags related or contradicting memories on save instead of silently piling up duplicates.
- You're in control — a built-in inspect UI to browse, edit, tag, import, and export everything you've stored.
Status
Early development. v1 is a local MCP server backed by SQLite, with hybrid (keyword + semantic) recall and local embeddings — nothing leaves your machine. See docs/plans for the design.
Tools
| Tool | Purpose |
|---|---|
recall |
Search your memory for relevant entries |
save_memory |
Store a new memory |
list_memories |
Browse stored memories |
update_memory |
Edit an existing memory |
forget |
Delete a memory |
Development
npm install
npm run build
npm start
The server speaks over stdio. Point an MCP client (e.g. Claude Code or Claude Desktop) at the built dist/index.js to connect:
claude mcp add engram -- node C:\Dev\Engram\dist\index.js
Inspect UI
A local web view to read, edit, tag, and export your memories lives in ui/:
cd ui
npm install
npm run dev
It reads the same store at ~/.engram/memory.db. Memories added or edited here are re-embedded by the server on its next start.
Storage
Everything lives in ~/.engram/ — memory.db (SQLite, source of truth) and models/ (the local embedding model). Set ENGRAM_DIR to use a different location.
License
MIT © Kalyan Gopalam
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。