Skip to content
Scribiz
Get transcript

Instagram Transcript API

Extract spoken voiceover, burned-in Reel captions, summaries and chapters from Instagram Reels via file upload or public link over one REST API.

POSThttps://scribiz.com/api/v1/context

Quickstart

Speech and Reel captions in two calls

Instagram blocks most server URL fetches with a login wall, so upload a saved Reel MP4 to /api/v1/uploads (up to 100 MB), then start a job on /api/v1/context with mode: "both" (or pass a public URL when available).

Safe to retry: Idempotency-Key prevents double charges
Terminal (upload MP4 + poll)
UPLOAD=$(curl -s -X PUT https://scribiz.com/api/v1/uploads \  -H "Authorization: Bearer $SCRIBIZ_API_KEY" \  -H "Content-Type: video/mp4" \  -H "X-File-Name: reel.mp4" \  --data-binary @reel.mp4 | jq -r .upload_id)[ -z "$UPLOAD" ] || [ "$UPLOAD" = "null" ] && exit 1JOB=$(curl -s https://scribiz.com/api/v1/context \  -H "Authorization: Bearer $SCRIBIZ_API_KEY" \  -H "Content-Type: application/json" \  -H "Idempotency-Key: $(uuidgen)" \  -d "{\"upload_id\": \"$UPLOAD\", \"mode\": \"both\"}" | jq -r .job_id)[ -z "$JOB" ] || [ "$JOB" = "null" ] && exit 1while true; do  STATE=$(curl -s "https://scribiz.com/api/v1/jobs/$JOB" \    -H "Authorization: Bearer $SCRIBIZ_API_KEY")  STATUS=$(printf '%s' "$STATE" | jq -r .status)  case "$STATUS" in    succeeded) break ;;    failed|canceled) printf '%s' "$STATE" | jq .error; exit 1 ;;  esac  sleep 3doneprintf '%s' "$STATE" | jq '{segments: .result.transcript.segments, scenes: .result.visual.scenes}'
Structured JSON responseupload_id + mode: "both"
GET /api/v1/jobs/{job_id}
{  "job_id": "job_8KX2nQpLm9RtVw4YsZaBcD",  "result": {    "meta": {      "durationSeconds": 18,      "title": "Day 1 of making a full dinner set"    },    "source": {      "fileName": "reel.mp4",      "kind": "file"    },    "synthesis": {      "summary": {        "tldr": "A potter throws the first bowl of an eight-piece dinner set and explains keeping only the best three."      }    },    "transcript": {      "language": "en",      "segments": [        {          "end": 17.4,          "start": 12,          "text": "Eight bowls, and I only like three of them. That is the whole process."        }      ]    },    "visual": {      "scenes": [        {          "description": "Potter centres wet clay on a wheel.",          "end": 6,          "onScreenText": [            "Day 1 of making a full dinner set"          ],          "start": 0,          "timingApprox": true        },        {          "description": "Finished bowl drying on a wooden shelf.",          "end": 12,          "onScreenText": [            "Bowl 1 of 8"          ],          "start": 6,          "timingApprox": true        }      ]    }  },  "status": "succeeded"}
Full reference: API endpointsOpenAPI: openapi.jsonBuilding an AI agent? Connect the MCP server

Audio + visual frame OCR vs caption-only scrapers

What a caption scraper misses on Instagram Reels: burned-in overlays and silent clips.

Caption-only scrapers and Scribiz Instagram Transcript API compared
Burned-in Reel captionsCaption-only scrapersReturns nothing on trending-audio Reels where the whole message is text on screenScribiz APImode: "both" or "watch" extracts on-screen headlines and captions with approximate scene times
Spoken voiceoverCaption-only scrapersInstagram rarely exposes clean downloadable subtitle tracksScribiz APITranscribes the Reel audio track itself, timed to about 0.2 seconds
Login wallsCaption-only scrapersReturns an empty result with no clear reason behind Instagram login wallsScribiz APIUpload the MP4 or audio (up to 100 MB) via PUT /api/v1/uploads; URL calls return source_auth_required clearly
Structured JSONCaption-only scrapersNo hook or scene breakdown beyond raw textScribiz APIReturns transcript.segments, visual.scenes, summary and chapters in one response
Agents & CLICaption-only scrapersSeparate tooling for web review, CLI scripts and AI assistantsScribiz APISame account works across REST API, CLI, web reader and remote MCP server

Evaluating developer APIs? See our side-by-side comparison on Scribiz vs Supadata.

Choose speech, screen OCR, or both per request

mode: "both"

2 min / video min

Transcribes spoken audio and reads burned-in hooks, overlays, slides and end cards from the frames onto one timeline.

mode: "listen"

1 min / video min

Spoken transcript with timestamps, speakers when there is more than one, summary and chapters when you do not need frame OCR.

mode: "watch"

1 min / video min

Reads only on-screen text and visual scene notes. For silent tutorials, recipes and music-only hooks.

Every free account starts with 30 minutes a month across the REST API, CLI, web app and MCP server. Opening a stored result again is free. See pricing and sources and limits.

Questions & answers

Common questions about the Scribiz Instagram Transcript API.

  • Instagram requires a login for most posts, so the reliable API flow is to upload the Reel MP4 or audio (up to 100 MB) with PUT /api/v1/uploads and pass upload_id to POST /api/v1/context with mode: "both" or "listen". You can also pass a public Reel URL, though hosted fetches are often refused with source_auth_required.
  • Yes. Many Reels pair trending music with burned-in text. Setting mode: "both" or "watch" returns result.visual.scenes with approximate scene times, description and onScreenText.
  • Instagram requires a login for most posts, so a hosted Reel URL is usually refused with source_auth_required and a hint to upload the file. Stories and private accounts never work by URL. Upload the saved MP4 via PUT /api/v1/uploads instead.
  • Listen uses 1 minute per minute of video and Both uses 2 minutes per minute. A 30-second Reel uses about 1 minute in Both mode. Free accounts include 30 minutes a month, and opening a stored result again is free.
  • Yes. Connect https://scribiz.com/mcp in Claude, Cursor, VS Code or Codex with your API key so agents can inspect videos with get_video_context, search_video and get_transcript.

Start with 30 free minutes.

Get a free API key in your dashboard or connect the remote MCP server inside Claude, Cursor or Codex.