Fovea Memory MCP
Local MCP server that enables AI clients to read and search memories from the Fovea macOS app, with privacy-focused local operation and optional semantic search.
README
Fovea Memory MCP
Fovea Memory MCP is a local stdio MCP server that lets AI clients read memories created by the Fovea macOS app.
It runs on your Mac, reads the local Fovea Memory SQLite database, and exposes only memories marked visible to MCP. It does not upload your memory database to Fovea, GitHub, npm, or any third-party service.
Install
Use this configuration in any MCP client that supports local stdio servers:
{
"mcpServers": {
"fovea-memory": {
"command": "npx",
"args": ["-y", "fovea-memory-mcp@1"]
}
}
}
The equivalent command is:
npx -y fovea-memory-mcp@1
No environment variables are required for normal use. The server discovers the current Fovea namespace through:
~/Library/Application Support/Fovea/current-namespace.json
Requirements
- macOS with Fovea installed and run at least once
- Node.js 22.5 or newer
- A local Fovea Memory database with memories visible to MCP
- An MCP client that supports local stdio servers
Client Examples
Claude Code
claude mcp add fovea-memory -- npx -y fovea-memory-mcp@1
Codex
codex mcp add fovea-memory -- npx -y fovea-memory-mcp@1
Claude Desktop, Cursor, VS Code, and Other Clients
Use the standard local server configuration:
{
"mcpServers": {
"fovea-memory": {
"command": "npx",
"args": ["-y", "fovea-memory-mcp@1"]
}
}
}
Some clients use a different top-level key, such as servers instead of mcpServers. Keep the server command and args the same.
Tools
The server exposes these MCP tools:
recall_recent_fovea_memory: recall what the user recently saw, asked, or said through Fovealist_recent_fovea_memories: list recent visible Fovea memoriessearch_fovea_memory: search visible Fovea memories by keyword, with optional semantic search when configuredforget_fovea_memory: soft-delete a memory from the local recent store
Privacy Model
- Local stdio process only
- Reads the local Fovea Memory SQLite database
- Returns only active memories with
mcp_visibility='visible' - Soft delete uses local SQLite status updates
- Writes best-effort local audit events to
audit.sqlite - Does not send memory data to GitHub or npm
Semantic search is optional. If FOVEA_BACKEND_URL and FOVEA_AUTH_TOKEN are set, the server can request an embedding for the search query. If they are not set, search automatically falls back to local keyword matching.
Development
npm test
The test suite creates a synthetic SQLite database at runtime. It does not require private Fovea fixtures.
License
Apache-2.0
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。