Plaud MCP & CLI: Use Voice Recordings from Claude & ChatGPT
A guide to Plaud's official MCP server (@plaud-ai/mcp) and CLI: search recordings, pull transcripts, and read AI summaries directly from Claude, ChatGPT, Cursor, and other AI clients. Covers one-command setup, the 7 MCP tools and 6 skills, terminal commands, meeting-to-follow-up workflows, and privacy considerations — based on the official docs.
Plaud MCP is the official MCP server (@plaud-ai/mcp) that connects recordings, transcripts, and AI summaries from PLAUD NOTE and NotePin voice recorders to any MCP-compatible AI client — Claude, ChatGPT, Cursor, Codex, and more. A companion Plaud CLI (@plaud-ai/cli) brings the same data to the terminal. 'Summarize Tuesday's standup' or 'draft a follow-up email from this meeting's action items' now work without opening the Plaud app. This guide is based on the official documentation.
From Recorder App to AI Workflow
Plaud is known for its card-shaped PLAUD NOTE and wearable NotePin AI voice recorders, with transcription and AI summaries delivered through its app. The MCP/CLI release opens that data to the context where users actually work — their AI agents. MCP (Model Context Protocol) is the standard that lets any compliant client connect to the same server.
The 7 MCP Tools
| Tool | What it does |
|---|---|
login / logout | Browser OAuth sign-in / revoke |
get_current_user | Account details |
list_files | List recordings with query and date filters |
get_file | Full details incl. a 24-hour presigned audio URL, speaker-labeled segments, AI notes |
get_note | AI summary, action items, key topics |
get_transcript | Full transcript with timestamps and speakers |
Six bundled skills automate common workflows: plaud-browse, plaud-find, plaud-read, plaud-digest (weekly reports), plaud-followup (follow-up emails and action items), and plaud-export (save to Notion, post to Slack). They load at install time and trigger from natural language.
One-Command Install
npx -y @plaud-ai/mcp@latest install
| Client | Auto-configured | Apply |
|---|---|---|
| Claude Desktop | ✓ | Quit and reopen |
| Claude Code | ✓ | New session (or /plugin install plaud) |
| Codex Desktop | ✓ | Quit and reopen |
| Cursor / Windsurf / VS Code / Zed | ✓ | Reload per client |
| Claude Web / ChatGPT Web | Interactive guide | Connect as connector/app |
Requirements are just Node.js 20+ and a Plaud account; --yes and --no-login flags cover CI and headless setups. For Claude Code MCP basics, see our Claude Code MCP integration guide.
Plaud CLI for Terminal and Automation
| Command | Purpose |
|---|---|
plaud login / logout / me | Auth |
plaud files / recent --days 30 / today | Browse recordings |
plaud search "Q2" --from 2026-04-01 --to 2026-04-30 | Keyword search (500 most recent) |
plaud transcript <id> -o out.txt | Transcript to file |
plaud summary <id> -o out.md | AI summary to file |
plaud audio <id> | 24-hour audio download URL |
Privacy Notes
- Web clients (Claude Web / ChatGPT Web) route recording data through Plaud's US-hosted MCP server — processed in transit only per the docs, but check your data-handling policies for sensitive meetings
- Presigned audio URLs live for 24 hours and remain in chat logs — mind the sharing scope
- Recording-consent obligations are unchanged
- Tokens live in ~/.plaud/; log out and remove it on shared machines
Is Plaud MCP free?
The MCP server and CLI are free npm packages requiring Node.js 20+ and a Plaud account. Transcription and AI summaries depend on your Plaud plan.
Which AI clients are supported?
Claude Desktop, Claude Code, Codex Desktop, Cursor, Windsurf, VS Code, and Zed are auto-configured; Claude Web and ChatGPT Web connect as a connector/app; any other MCP client can add the npx command manually.
Where does my recording data travel?
Local clients run the MCP server locally. HTTP-based web clients route through Plaud's US-hosted server, processed in transit only per the docs — verify against your internal policies for confidential content.
Takeaway
Plaud MCP/CLI breaks the assumption that recorder data lives only in the vendor's app, placing meeting knowledge directly in your AI agent's working context. Setup is effectively one command, and a hardware vendor shipping an official MCP server is itself a sign of where 2026 workflows are heading: record the meeting, then just ask your AI.
Feel free to contact us
Contact Us