Your Plaud Notes contain private conversations, meetings, and personal recordings. Before connecting them to Claude, follow this guide to lock down your Claude account so your data stays private.
Do this BEFORE setting up the Plaud Notes connection.
By default, Claude may use your conversations to improve its models. Turn this off.
- Go to claude.ai/settings
- Click "Data Privacy Controls" (or go directly to claude.ai/settings/data-privacy-controls)
- Find the "Model Improvement" toggle
- Turn it OFF
What this does: With training OFF, your conversations are kept for only 30 days (instead of an extended period) and are never used to train Claude's models. Your meeting transcripts and recordings won't become part of Claude's training data.
Claude's Memory feature remembers things from your past conversations, which can be very useful -- it lets Claude build context about your projects, clients, and preferences over time.
- Go to claude.ai/settings
- Click "Capabilities" (or "Memory" if you see it)
- Confirm that Memory is ON (this is the default and recommended for most users)
- Review your saved memories periodically by clicking "Manage Memory" and remove any specific memories you don't want kept
Why keep it on: Memory is what makes Claude your second brain. It lets Claude remember your preferences, project details, and context across conversations. If Claude remembers that Client X prefers a specific tech stack, or that you discussed a deadline last week, that's Memory working for you.
Tip: You can always delete individual memories you don't want stored. Go to Settings > Memory > Manage Memory to review and remove specific items.
Claude has a Temporary Chat mode that creates conversations that are never saved and never used for training.
- When starting a new conversation about sensitive topics, toggle on "Temporary Chat" before sending your first message
- Use Temporary Chats whenever you're asking Claude to pull up meeting transcripts or especially sensitive recordings
What this does: Temporary conversations are not stored in your chat history, not used for training, and not included in Claude's memory -- regardless of your other settings.
- If you use email login: Make sure you have a strong, unique password for your Claude account
- Better option: Sign in with Google (with Google's 2-Step Verification enabled) or Apple (with Apple's two-factor authentication) -- this gives you MFA protection that Claude's email login doesn't offer on its own
- Best option (for teams): If you're on a Claude Team or Enterprise plan, set up SSO (Single Sign-On) through your company's identity provider (Okta, Google Workspace, etc.) which enforces multi-factor authentication
When Claude connects to your Plaud Notes server, it will ask for permission to use each tool. Be intentional about what you approve.
- When Claude asks to use a Plaud Notes tool for the first time, review what it's doing before clicking "Allow"
- Don't click "Allow always" unless you fully trust the setup -- approve each time instead, especially at first
- To disable tools you don't need: click the "Search and tools" menu in Claude Desktop and toggle off any Plaud tools you're not currently using
- Don't use
--dangerously-skip-permissions-- this bypasses all safety prompts - Consider using
--sandboxmode for extra isolation:claude --sandbox
If you use Claude Code (the CLI tool), add these to your shell profile (~/.bashrc, ~/.zshrc, or similar):
# Disable non-essential data collection
export CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1
export DISABLE_TELEMETRY=1
export DISABLE_ERROR_REPORTING=1- Add the lines above to your shell profile
- Restart your terminal or run
source ~/.bashrc(or~/.zshrc)
What this does: Stops Claude Code from sending operational metrics, error reports, and surveys. None of these include your code or file contents, but disabling them is good practice for sensitive work.
The /feedback command in Claude Code sends your entire conversation transcript to Anthropic with an extended retention period.
- Do not use
/feedbackin any session where you've loaded Plaud Notes transcripts or meeting content
After you're done working with sensitive Plaud Notes data:
- Delete the conversation from your Claude chat history (click the three dots on the conversation, then "Delete")
- Deleted conversations are purged from Anthropic's servers within 30 days and are excluded from any future training
Tip: If you used an Temporary Chat (Step 3), there's nothing to delete -- it was never saved.
Understanding what data goes where helps you make informed decisions:
| Data | Where It Lives | Who Can See It |
|---|---|---|
| Your Plaud recordings (audio files) | Plaud's servers (plaud.ai) | You + Plaud |
| Your Plaud token | Your machine / your cloud host | You only |
| MCP server (this project) | Your machine or your cloud host | You only |
| Tool outputs (transcript text sent to Claude) | Anthropic's servers (30 days with training OFF) | Anthropic's automated systems; human review only if flagged for safety |
| Claude's responses | Anthropic's servers (30 days with training OFF) | Same as above |
| Temporary chat content | Not stored | No one after session ends |
Key insight: Your raw audio files never leave Plaud's servers or your MCP server. Only the text that the MCP tools return (transcripts, summaries) gets sent to Anthropic's API for Claude to process.
If your Plaud recordings contain business-sensitive information (client meetings, legal discussions, etc.), consider upgrading to a Claude plan with stronger data protections:
| Feature | Free/Pro/Max | Team | Enterprise |
|---|---|---|---|
| Opt out of training | Manual toggle | Off by default | Off by default |
| Data retention | 30 days (training off) | 30 days | Custom (min 30 days) |
| SSO with MFA | No (use Google/Apple 2FA) | Yes | Yes |
| Zero Data Retention | No | No | Available by arrangement |
| Admin MCP controls | No | Limited | Full (managed-mcp.json) |
| Custom data policies | No | No | Yes |
For the highest level of protection, Enterprise customers can request Zero Data Retention (ZDR) -- Anthropic will not store your inputs or outputs at all.
Before connecting Plaud Notes to Claude, confirm you've done all of these:
- Training toggle is OFF (claude.ai/settings/data-privacy-controls)
- Memory is ON and you know how to manage/delete individual memories
- You know how to start Temporary Chats
- Your login uses MFA (Google 2FA, Apple 2FA, or SSO)
- You understand MCP tool permissions and won't auto-approve blindly
- Claude Code privacy env vars are set (if using CLI)
- You will never use /feedback in sensitive sessions
- You will delete conversations with sensitive data when done
Once all boxes are checked, proceed to the Setup Guide to connect your Plaud Notes.