| name | plaud-read |
| version | 1.0.0 |
| description | Read the transcript, AI summary, notes, or download audio for a specific Plaud recording. Use when the user says 'show the transcript', 'summarize this', 'what was said', 'get audio', 'the notes from', or names a specific recording to dig into. Also covers extracting structured fields from a recording. |
| metadata | {"requires":{"bins":[]}} |
plaud-read
Read plaud-shared first.
When to use
- User names a specific recording (by name or ID) and wants to read its content.
- User asks for "transcript", "summary", "action items", "audio", "who said what", or a structured extraction ("action items, decisions, attendees").
- If the user did not specify a recording, hand off to
plaud-find (by topic) or plaud-browse (by recency) first.
Tool selection matrix
| User wants | Tool | Notes |
|---|
| AI summary, TL;DR, action items | get_note | Returns Markdown; usually enough — try this before get_transcript |
| Verbatim quotes, full dialogue | get_transcript | Timestamped; larger |
| Audio download link | get_file then use presigned_url | Link expires in 24h |
| Full metadata + availability flags | get_file | Check source_list / note_list populated before claiming content exists |
Structured extraction workflow
If the user provides a schema (e.g., {"action_items": [], "decisions": [], "attendees": []}):
- Call
get_note first — the AI summary usually already contains these fields.
- Only call
get_transcript if the summary is missing a required field.
- Return JSON matching the user's schema. Mark any missing field with
null and note why.
Common schemas:
- Sales:
{ "pain_points": [], "follow_ups": [], "deal_stage": "" }
- Clinical:
{ "diagnoses": [], "medications": [], "next_appointment": "" }
- Project:
{ "action_items": [], "decisions": [], "attendees": [] }
Output
- Transcripts: preserve
[MM:SS - MM:SS] Speaker: content. Do not reformat timestamps.
- Summaries: render Markdown directly in the reply.
- Audio: print the URL and mention "expires in 24h".
Anti-patterns
- Do not call
get_transcript speculatively — it's the largest payload.
- Do not paraphrase the AI summary unless the user asked; quote it verbatim.
