sovseal MCP Server
Provides a private, on-device memory with semantic search for MCP clients, enabling encrypted storage and recall of context facts without external API calls.
README
sovseal MCP Server
An MCP server that gives Claude, Cursor, and every MCP client one shared memory that never leaves your machine. On-device vector search via LanceDB + Transformers.js (384-dim embeddings), AES-256-GCM client-side encryption, and write-behind ciphertext sync. 0 RTT reads.
<a href="https://glama.ai/mcp/servers"><img width="380" height="200" alt="sovseal MCP server" src="https://glama.ai/mcp/servers/badge" /></a>
Tools
Store Memory (store_memory)
Embeds and persists a context fact inside the local on-device memory node.
Parameters:
content(string, required): A third-person factual statement to store (max 65,536 chars)
Behavior:
- Embeds content on-device (intfloat/multilingual-e5-small, 384-dim, ONNX)
- Writes to local LanceDB — returns immediately
- Ciphertext sync runs write-behind; nothing blocks the caller
- Automatically de-duplicates and reinforces existing entries
- High-risk PII (SSNs, credit cards, API keys) is automatically redacted before storage
Recall Memory (recall_memory)
Semantic search over stored memory context. Returns top-K relevant memories by L2 distance.
Parameters:
query(string, required): Semantic query string to search past memoriestopK(number, optional): Maximum number of memories to return (1–20, default: 5)
Behavior:
- Embeds query on-device (LRU-cached pipeline) → vector search local LanceDB
- 0 RTT — network is never on the read path
- Returns
[score=NUMBER id=ID] textfor each match — smaller score = closer semantic match - Returns
no_matcheswhen the store has no relevant entries
Resources
Briefing Context (sovseal://context/briefing)
A summarized, reinforcement-aware briefing of procedural, semantic, and recurring memories. Recommended for bootstrapping new conversations with stored user context.
Recent Context (sovseal://context/recent) — Deprecated
Raw list of the most recently stored context facts. Use sovseal://context/briefing instead.
Prompts
Context Injection (/sovseal:context)
System prompt helper that bootstraps a conversation by reading from sovseal://context/recent before the first turn.
Configuration
No API Key Required
sovseal runs 100% on-device. There is no external API key, no cloud dependency, and no account required for local memory operations.
Environment Variables
The server supports the following environment variables:
SOVSEAL_TRANSPORT: Transport mode ("stdio"or"sse", default:"stdio")SOVSEAL_PORT: SSE server port (default:4040)
Installation
Usage with Claude Desktop
Add this to your claude_desktop_config.json:
{
"mcpServers": {
"sovseal-memory": {
"command": "npx",
"args": ["-y", "@sovseal/mcp-server"]
}
}
}
Usage with Cursor
Add to your Cursor MCP settings (~/.cursor/mcp.json):
{
"mcpServers": {
"sovseal-memory": {
"command": "npx",
"args": ["-y", "@sovseal/mcp-server"]
}
}
}
Usage with VS Code (Copilot / Cline / Roo)
Add to your User Settings (JSON) or .vscode/mcp.json:
{
"servers": {
"sovseal-memory": {
"command": "npx",
"args": ["-y", "@sovseal/mcp-server"]
}
}
}
Usage with Windsurf / Zed / OpenCode
{
"mcpServers": {
"sovseal-memory": {
"command": "npx",
"args": ["-y", "@sovseal/mcp-server"]
}
}
}
Automatic Onboarding (All Clients)
Detect your IDE, configure global connection settings, and write system instructions automatically:
npx -y @sovseal/mcp-server onboard --write --register
Supported: Cursor, Claude Desktop/Code, Windsurf, VS Code (Copilot/Cline/Roo), Zed, Google Antigravity, and OpenCode.
HTTP/SSE Transport
For always-on autonomous agents, run as a long-lived HTTP/SSE server:
SOVSEAL_TRANSPORT=sse SOVSEAL_PORT=4040 npx -y @sovseal/mcp-server
Hosted Endpoint
A hosted HTTP/SSE endpoint is available for web-based MCP clients and marketplace validation:
https://sovseal-mcp-server.barrackbobby1.workers.dev/mcp
How It Stays Private
- AES-256-GCM client-side encryption with a 96-bit random IV per snapshot
- 256-bit key lives in
~/.sovseal/config.json(mode0600) — lose this file, lose every snapshot - Server only ever sees ciphertext + SHA-256-derived paths; it cannot read your context
- Verified Semantic Recall (VSR) — every load re-derives
sha256(canonicalize(payload))and fails closed on mismatch - Packet-Capture Guarantee — run Wireshark or tcpdump against the server; if you find a single byte of unencrypted context on the wire, the software is free forever
Performance
| Workload | Operation | p50 | p95 | p99 |
|---|---|---|---|---|
| 10K records · 1K queries | recall_memory (warm) |
6.1 ms | 10.4 ms | 21.8 ms |
| Single write | store_memory |
3.8 ms | 7.2 ms | 12.5 ms |
| First call | recall_memory (cold) |
~1.2 s | — | — |
All operations are 0 RTT — network is never on the read path.
Build
npm install
npm run build
Development
Prerequisites
- Node.js 20.x or higher
- npm or pnpm
Setup
- Clone the repository:
git clone https://github.com/sovseal/mcp-server.git
cd mcp-server
- Install dependencies:
npm install
- Build the project:
npm run build
Testing via MCP Inspector
- Build and start the server:
npm run build
node dist/index.js
- In another terminal, start the MCP Inspector:
npx @modelcontextprotocol/inspector node dist/index.js
Available Scripts
npm run build: Build the TypeScript projectnpm run dev: Watch for changes and rebuildnpm run typecheck: Type-check without emittingnpm run test: Run test suite (Vitest)npm run test:watch: Watch mode testingnpm run bench: Run performance benchmarks
Links
- Website: sovseal.com
- Documentation: docs.sovseal.com
- Trust Center: trust.sovseal.com
- Chrome Extension: Chrome Web Store
- Full Monorepo: github.com/sovseal/core
License
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。