plaud-mcp
Enables querying Plaud AI recordings, transcripts, and summaries through the Plaud Desktop app's authenticated session via Chrome DevTools Protocol.
README
Plaud MCP Server
MCP server for Plaud transcripts via CDP proxy through the running Plaud Desktop app.
How It Works
Claude Code → MCP Server → CDP (WebSocket) → Plaud Desktop → Plaud API
The MCP connects to the running Plaud Desktop Electron app via Chrome DevTools Protocol:
- Sends
SIGUSR1to the Plaud Desktop process to enable Node.js inspector (port 9229) - Connects via WebSocket to the inspector
- Executes API calls through the app's own authenticated
$fetchfunction - Returns results back through MCP tools
No token extraction, no cookies, no API keys. Uses the app's live authenticated session directly.
Prerequisites
- Plaud Desktop - installed and signed in
- Python 3.10+
- uv (recommended)
Installation
# Install as a CLI tool (recommended)
uv tool install "plaud-mcp @ git+https://github.com/davidlinjiahao/plaud-mcp"
# Or install from local source
uv tool install --force "plaud-mcp @ ."
Configuration
Add to ~/.claude.json:
{
"mcpServers": {
"plaud": {
"command": "${HOME}/.local/bin/plaud-mcp",
"args": [],
"env": {}
}
}
}
No API keys or tokens needed. Just ensure Plaud Desktop is running.
MCP Tools
| Tool | Description |
|---|---|
check_connection |
Verify Plaud Desktop is available |
get_file_count |
Total number of recordings |
get_recent_files |
Files from the last N days |
get_files |
Files with optional date filters |
get_file |
Metadata for a specific file |
get_transcript |
Full transcript with speaker labels |
get_summary |
AI-generated summary |
search_transcripts |
Search transcripts by content |
Troubleshooting
"Plaud Desktop is not running"
Launch the Plaud Desktop app and sign in.
"Could not enable inspector"
The SIGUSR1 signal may have failed. Ensure Plaud Desktop is the main process, not a helper.
Search is slow
search_transcripts fetches and searches client-side. Reduce the days parameter.
Why CDP?
Plaud's API validates auth at the Chromium network stack level - tokens extracted from LevelDB don't work with standard HTTP clients (httpx, curl, curl_cffi with Chrome impersonation all return 401). The CDP approach bypasses this entirely by executing requests through the app's own authenticated context.
Development
uv venv && source .venv/bin/activate
uv pip install -e ".[dev]"
ruff check src/ && pyright src/
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。