Plaud MCP Server

Plaud

Plaud Note and Plaud NotePin are hardware AI voice recorders. Recordings sync from the device to the user's Plaud cloud account. The cloud runs the audio through speech-to-text, plus an LLM, to produce a transcript, an automatic summary, and a list of action items. The user can also write a free-form note alongside the AI summary.

This bundle is a read-only view of one Plaud account. It cannot upload, delete, transcribe, or modify recordings. It cannot create or edit notes, summaries, or device entries. The web app at web.plaud.ai and the iOS / Android Plaud apps remain the place to do all of that.

How a Plaud account is shaped

• User — one account, identified by email. Has a region (the AWS region the account was created in), a membership tier (free / Pro / Unlimited), monthly transcription quota, and a Stripe subscription if paid.
• Devices — Plaud Notes and NotePins paired to the account. Each carries a serial number, model, firmware version, and battery state.
• Recordings (file) — every captured audio clip. Carries: filename, duration in seconds, capture date, file size, source device serial, "scene" tag (meeting / interview / lecture / …), and compact availability fields for transcript, summary, and hand-edited notes.
• AI artifacts (per recording) — every AI-generated derivative of the audio is an artifact under that recording's detail response. The two confirmed artifact families are:
  • auto_sum_note — the auto-generated markdown summary. Inlined as text in pre_download_content_list[].data_content.
  • consumer_note — the user's hand-edited markdown note for that recording (free-form, optional). Inlined the same way.
  • source_transaction — a binary blob that backs the raw transcript pipeline. Useful as a pointer; the human-readable transcript is the inline trans_result segments on the recording itself.

pre_download_content_list[] is the artifact list with content already in the response (data_content is the actual markdown). content_list[] is the parallel list with just metadata + a pre-signed S3 link (data_link). The dedicated transcript, summary, and note tools resolve those shapes so agents do not need to inspect raw artifact rows.

Each artifact id has the shape {type}:{short_hash}:{file_id} so an artifact's family can be read off the prefix without joining to anything.

Workflow order

1. Confirm the connected account first: email, plan, transcription quota, and region.
2. Inspect paired devices to understand which Plaud Notes or NotePins are contributing recordings.
3. List recordings to get compact metadata and content availability, then drill into a single recording only when the user asks what content is available for that file.
4. Use the transcript view for timestamped {start_time, end_time, content, speaker} segments.
5. Use the summary view for the auto-generated markdown summary from the auto_sum_note artifact.
6. Use the user-note view for the hand-edited markdown note from the consumer_note artifact. It is empty when the user has not written one.
7. Request the original-audio download URL only when the user explicitly needs the audio file. The URL still requires the connected Plaud credential, but the tool does not expose bearer tokens.
8. Poll AI processing status when a recording was just uploaded and the transcript or summary has not appeared yet.

Region routing is automatic

Plaud runs separate regional API servers (api-euc1, api-use1, api-apse1, …) and the JWT carries which region the user belongs to. Tools always go to the correct regional host. There is nothing to configure.

What this bundle does not do

• It does not upload audio. The hardware device + the official Plaud apps own that path.
• It does not delete recordings. Use the Plaud app.
• It does not regenerate summaries with a different prompt or template. Use the Plaud app.
• It does not see other people's recordings. Access is scoped to one Plaud account.