Plaud Notes MCP Server
Connects your Plaud voice recorder to Claude, giving access to recordings, transcripts, and AI summaries so you can query your notes via natural language.
README
Plaud Notes MCP Server
Connect your Plaud voice recorder to Claude. Claude gets access to your recordings, transcripts, and AI summaries -- so you can ask it about anything you've ever recorded.
New here? No coding experience? Follow the Step-by-Step Setup Guide -- it walks you through everything with checkboxes.
Before you start: Read the Security Setup Guide to lock down your Claude account first. Your Plaud Notes contain private conversations -- make sure they stay private.
<!-- To enable one-click deploy: register a template at railway.com/button,
then replace TEMPLATE_ID below with your template slug -->
What Can Claude Do With Your Plaud Notes?
Once connected, just talk to Claude naturally:
- "Show me my recent recordings"
- "Get the transcript from my meeting yesterday"
- "Search my notes for anything about the Q4 budget"
- "Summarize my last 3 recordings"
- "What did we discuss about hiring?"
- "Load my last 5 recordings for context, then help me write a follow-up email"
Features
- Browse Recordings - See all your notes with titles, durations, and dates
- Full Transcripts - Complete transcriptions with speaker labels
- AI Summaries - Plaud's AI-generated summaries and notes
- Search Everything - Find any topic across all your recordings
- Bulk Context Loading - Pull recent transcripts into Claude in one go
- Tags, Speakers, Devices - Filter by folder, see who spoke, check your devices
Quick Start
The short version (3 steps):
- Get your token from web.plaud.ai (DevTools → Local Storage →
tokenstr) - Deploy this server (click the Railway button above, or use Docker)
- Point Claude at it
Detailed instructions below, or see the full beginner guide.
1. Get Your Plaud Token
- Sign in to web.plaud.ai
- Open browser DevTools (
F12orCmd+Option+I) - Go to Application tab (or Storage in Firefox) → Local Storage →
https://web.plaud.ai - Find
tokenstrand copy its value - The token is the part after
bearer(starts witheyJ...)
Tip: Your token lasts ~10 months. Alternatively, go to Application > Local Storage and copy the
tokenstrvalue.
Note: If you use Google or Apple sign-in, you'll need to set a password first via "Forgot Password" on web.plaud.ai to ensure token compatibility.
2. Install
git clone https://github.com/jameshenning/PlaudNotes.git
cd PlaudNotes
# Option A: Install directly
pip install -e .
# Option B: Use a virtual environment (recommended)
python3 -m venv venv
source venv/bin/activate # Windows: venv\Scripts\activate
pip install -e .
3. Configure Your Token
Choose one method:
# Environment variable (simplest)
export PLAUD_TOKEN="eyJhbGciOiJI..."
# Or .env file
cp .env.example .env
# Edit .env and paste your token
# Or config file (most persistent)
mkdir -p ~/.config/plaud
echo "eyJhbGciOiJI..." > ~/.config/plaud/token
chmod 600 ~/.config/plaud/token
EU region users - set the region or API domain:
export PLAUD_REGION=eu
# or explicitly: export PLAUD_API_DOMAIN=https://api-euc1.plaud.ai
The server auto-detects region mismatches and redirects automatically, but setting the correct region avoids the extra round-trip.
4. Connect to Claude
Claude Code (CLI):
# Basic setup
claude mcp add plaud-notes -- python -m plaud_notes_mcp.server
# With environment variable for token
claude mcp add plaud-notes -e PLAUD_TOKEN=your_token_here -- python -m plaud_notes_mcp.server
# With virtual environment
claude mcp add plaud-notes -e PLAUD_TOKEN=your_token_here -- /path/to/PlaudNotes/venv/bin/python -m plaud_notes_mcp.server
# Verify it's registered
claude mcp list
Claude Desktop:
Edit your config file:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"plaud-notes": {
"command": "python",
"args": ["-m", "plaud_notes_mcp.server"],
"cwd": "/path/to/PlaudNotes",
"env": {
"PLAUD_TOKEN": "your_token_here"
}
}
}
}
Fully quit and restart Claude Desktop after saving.
Remote Deployment (Docker)
Deploy the server to the cloud so Claude connects to it over HTTP -- no local install needed.
Security: Always set
PLAUD_MCP_API_KEYwhen deploying over HTTP. This requires all MCP clients to authenticate with a Bearer token, preventing unauthorized access to your recordings.Generate a key:
python -c "import secrets; print(secrets.token_urlsafe(32))"
Docker (self-hosted)
# Clone and configure
git clone https://github.com/jameshenning/PlaudNotes.git
cd PlaudNotes
cp .env.example .env
# Edit .env: set PLAUD_TOKEN, PLAUD_MCP_API_KEY, and optionally PLAUD_REGION
# Run with Docker Compose
docker compose up -d
# Server is now running at http://localhost:8000/mcp
Then connect Claude Code to the remote server:
# Without API key auth
claude mcp add --transport http plaud-notes http://localhost:8000/mcp
# With API key auth (pass as header)
claude mcp add --transport http -H "Authorization: Bearer YOUR_API_KEY" plaud-notes http://localhost:8000/mcp
Railway (cloud hosting)
- Go to railway.app and create a free account
- Click "New Project" → "Deploy from GitHub repo"
- Connect your GitHub account and select PlaudNotes (fork it first if you don't see it)
- Once Railway starts building, click the service → "Variables" tab
- Add these environment variables:
PLAUD_TOKEN= your Plaud token (without thebearerprefix)PLAUD_MCP_API_KEY= a secret password you make upPLAUD_TRANSPORT=httpPLAUD_MCP_HOST=0.0.0.0PLAUD_REGION=us(oreu)
- Click "Settings" → "Networking" → "Generate Domain"
- Railway auto-detects the
Dockerfileand deploys - Copy your Railway URL and connect Claude:
claude mcp add --transport http plaud-notes https://your-app.up.railway.app/mcp
Fly.io
# Install the Fly CLI, then:
cd PlaudNotes
fly launch --no-deploy
# Set secrets (never in fly.toml)
fly secrets set PLAUD_TOKEN="eyJhbGciOiJI..."
fly secrets set PLAUD_MCP_API_KEY="your_generated_key"
# Deploy
fly deploy
# Connect Claude to your Fly app
claude mcp add --transport http plaud-notes https://plaud-notes-mcp.fly.dev/mcp
Render / Generic Docker Host
Any platform that runs Docker containers works. Set these environment variables:
| Variable | Required | Default | Description |
|---|---|---|---|
PLAUD_TOKEN |
Yes | - | Your Plaud bearer token |
PLAUD_MCP_API_KEY |
Recommended | - | API key for HTTP auth (Bearer token) |
PLAUD_TRANSPORT |
Yes (remote) | stdio |
Set to http for remote deployment |
PLAUD_MCP_HOST |
No | 127.0.0.1 |
Bind address (set to 0.0.0.0 for Docker/cloud) |
PLAUD_MCP_PORT |
No | 8000 |
Listen port |
PLAUD_REGION |
No | us |
us or eu |
PLAUD_API_DOMAIN |
No | - | Override API URL directly |
Connect Claude Desktop to a Remote Server
For Claude Desktop, add the remote URL to your config:
{
"mcpServers": {
"plaud-notes": {
"type": "http",
"url": "https://your-deployment-url.example.com/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
}
}
}
}
Available Tools
Once connected, Claude has access to 12 tools:
| Tool | Description |
|---|---|
list_recordings |
Browse recordings with titles, durations, dates, and transcript/summary status |
get_transcript |
Get the full transcript of a recording with speaker labels |
get_summary |
Get the AI-generated summary and notes for a recording |
get_recording_detail |
Get complete metadata, transcript, and AI content for a recording |
search_notes |
Search across all recordings for specific content in titles, transcripts, or summaries |
get_recent_context |
Load transcripts and summaries from your N most recent recordings at once |
get_recordings_by_tag |
Get all recordings within a specific tag/folder |
list_tags |
List your organizational tags and folders |
list_speakers |
List all identified speakers across recordings |
get_audio_url |
Get a temporary pre-signed audio download URL |
get_user_info |
Check your Plaud account information and membership status |
list_devices |
List connected Plaud devices (NotePin, Note, Note Pro) |
Example Conversations
Once configured, you can ask Claude things like:
- "Load my last 5 Plaud recordings for context" - uses
get_recent_context - "Show me my recent Plaud Notes recordings" - uses
list_recordings - "Get the transcript from my meeting yesterday" - uses
get_transcript - "Search my notes for anything about the Q4 budget" - uses
search_notes - "Summarize the key points from my last 3 recordings" - uses
get_recent_context+ analysis - "What did we discuss about the product roadmap?" - uses
search_notes - "Show me all recordings tagged as Work" - uses
list_tags+get_recordings_by_tag - "Who were the speakers in my last meeting?" - uses
list_speakers
Troubleshooting
| Issue | Solution |
|---|---|
Authentication error |
Your token may be expired. Sign into web.plaud.ai again and get a new token. |
No recordings found |
Ensure "Private Cloud Sync" is enabled in the Plaud mobile app (Me > Private Cloud Sync). |
Region mismatch |
The server auto-redirects, but you can explicitly set PLAUD_REGION=eu if your account is in the EU. |
Connection refused (local) |
Verify the server is running. Test with python -m plaud_notes_mcp.server directly. |
Connection refused (remote) |
Check that PLAUD_TRANSPORT=http and PLAUD_MCP_HOST=0.0.0.0 are set. Check your cloud platform logs. |
| Docker container exits | Run docker compose logs to see the error. Usually a missing PLAUD_TOKEN. |
Architecture
PlaudNotes/
src/plaud_notes_mcp/
__init__.py # Package metadata
plaud_client.py # Plaud API client (auth, recordings, transcripts, summaries)
server.py # MCP server (12 tools, 3 resources, stdio + HTTP transport)
Dockerfile # Container image for remote deployment
docker-compose.yml # One-command local Docker setup
fly.toml # Fly.io deployment config
railway.toml # Railway deployment config
railway.json # Railway template config
Procfile # Generic PaaS entry point
pyproject.toml # Python package definition
.env.example # Configuration template
The server uses the Plaud web API (the same API that powers web.plaud.ai). This is a reverse-engineered, unofficial API -- it is not endorsed by Plaud and may change without notice.
Transport modes:
- stdio (default) -- for local use with Claude Code CLI and Claude Desktop
- HTTP (
PLAUD_TRANSPORT=http) -- for remote/cloud deployment, serves at/mcp
Security Notes
Important: Before connecting sensitive data, follow the Security Setup Guide to harden your Claude account (disable training, enable 2FA, review privacy settings).
Your data flow:
- Raw audio files stay on Plaud's servers -- they are never sent to Claude
- Only transcript text and summaries (returned by MCP tools) are sent to Claude's API
- With training OFF, Anthropic retains data for 30 days only and never trains on it
- Using Incognito Chats means nothing is stored at all
Server security:
- Your Plaud bearer token is stored locally and only sent to Plaud's own API servers
- When deployed over HTTP, set
PLAUD_MCP_API_KEYto require authentication - All
file_idinputs are validated against path traversal and injection attacks - API redirects are restricted to known Plaud domains only
- Docker container runs as non-root with read-only filesystem
- Never commit your
.envfile or token to version control (.gitignorecovers this)
Need Help?
- Security Setup Guide -- Lock down your Claude account before connecting
- Step-by-Step Setup Guide -- Beginner-friendly checklist
- Open an Issue -- Report bugs or ask questions
License
MIT
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。