mcps
D T E

Recipe

Turn one long recording into a clip pack

Find the moments by describing them, cut by editing the transcript, caption for free — and optionally re-voice a clip in another language with ElevenLabs.

for Creators who publish weekly to Instagram, LinkedIn, Twitter, Threads, Telegram

Hand-checked recipe · curated by Ilya Gindin

~60 minutes

to run

beginner

difficulty

3

connectors

What you'll need

Descript The engine: finds moments in the transcript, cuts by text, captions, exports verticals.
Required
Transcriptor Optional — pulls the transcript of an outside video (e.g. a YouTube link) so you can scout moments before importing.
Optional
ElevenLabs Optional — re-voices a clip in another language or a cleaner read for dubbing.
Optional

This recipe runs in 4 steps

  1. 1

    Find the moments by describing them

    Say to your AI

    In this recording, find the section where I explain pricing, the story about the first customer, and the hot take about tools.

    You'll get: Descript locates each moment in the transcript — no scrubbing.

  2. 2

    Cut a clip by editing text

    Say to your AI

    Cut the pricing section into a 45-second clip: remove filler words and the false start, keep the punchline.

    You'll get: A tight clip, edited the way you'd edit a paragraph.

  3. 3

    Caption and export the pack

    Say to your AI

    Add bold centered captions to each clip and export them all as 1080x1920 verticals.

    You'll get: A pack of captioned vertical clips from one recording.

  4. 4

    Optional: dub one for another market

    Say to your AI

    Take the pricing clip's text and read it in Spanish, warm and conversational — I'll lay it over the clip.

    You'll get: An ElevenLabs voice track to swap under the same visuals.

You're done when

One long recording becomes a pack of captioned vertical clips — found by describing moments, cut by editing text — plus an optional dubbed variant for another language.

The chain is Descript (+ optionally Transcriptor and ElevenLabs), about an hour for a full pack: you FIND moments by describing them in plain words, CUT them by editing the transcript like a document, and captions come out accurate because they’re the same text. The optional last hop — ElevenLabs re-voicing a clip in another language over the same visuals — is the highest-leverage trick in the chain.

This is the one-asset-to-many job every creator has, and the seam is invisible until you chain the tools: transcript-first editing means search, cut, and captions all share one source of truth. A timeline editor can’t offer that; the repos don’t mention each other.

Checked 2026-07-12: Descript official and reachable; Transcriptor community (optional); ElevenLabs official (optional).

If something breaks

I don't see Descript to turn on
Fix: In Claude: Customize → Connectors → search Descript → Connect (or add the custom URL https://api.descript.com/v2/mcp). In ChatGPT: add the Descript app.
It can't find the moment I described
Fix: Describe it by the words spoken, not the vibe: 'where I say churn doubled' beats 'the sad part'. The search runs on the transcript.
The transcript is rough in places
Fix: Noisy audio and heavy accents lower accuracy — proofread the transcript before cutting; the captions inherit whatever it says.
The dub doesn't match the clip length
Fix: Ask ElevenLabs for the same pacing ('match a 45-second read, conversational') and trim the clip tail in Descript rather than stretching audio.

People ask their AI

“cut my podcast into short clips”“repurpose a long video into shorts”“find the best moments in my recording”

Related

Descript ElevenLabs talking head explainer from a script