Descript

Category: AI Video

Edit video by editing the transcript. Watch the credit meter.

Descript edits video and audio as a document. It transcribes your recording, and deleting a word from the transcript removes it from the footage, which makes cutting a long interview dramatically faster than scrubbing a timeline. Around that sit the features that made it popular: filler-word removal in one click, Studio Sound for cleaning up bad audio, Overdub for correcting a misspoken word by typing the right one in your own cloned voice, eye contact correction, green screen without a green screen, screen recording, multitrack editing and a stock library. Underlord is the AI layer, handling clips, show notes, titles and edits from a prompt. There is a candidly documented first-party MCP server, hosted with OAuth and listed in both the ChatGPT and Claude directories, which can import media, run Underlord edits, publish and export transcripts and timelines to Premiere, Resolve and others.

Our verdict

Transcript-based editing remains one of the genuinely transformative ideas in creative software, and if you produce talking-head video or podcasts regularly it will save you hours a week from the first session. The MCP server is unusually honest, publishing its own limits openly, including that it cannot export MP4 and only publishes to a share link. That candour is rare and worth noting. The thing to go in knowing is the 2025 repricing, which is still the live grievance among existing users. Descript moved from transcription-hour billing with largely uncapped AI to a media-minute meter plus a separate metered AI credit pool, so headline prices stayed similar while what they buy shrank. Expect bill shock during a busy production month, and be aware that avatar and video generation drain the entry tier's credit pool fast. Their own documentation warns that automated editing through MCP burns credits faster than editing manually. Budget on your real monthly output, not the sticker price.

Best for: Marketers and podcasters producing regular video and audio who want to edit at the speed of writing rather than learning a timeline editor.

Pros

  • Transcript-based editing is dramatically faster than a timeline for talking-head content
  • Hosted MCP server with OAuth, in both the ChatGPT and Claude directories
  • Unusually honest published documentation of what the MCP server cannot do
  • Filler-word removal, Studio Sound and eye contact correction all work well
  • Exports timelines to Premiere, Resolve and Final Cut if you finish elsewhere
  • Free tier is enough to judge whether the workflow suits you

Cons

  • The 2025 repricing swapped generous AI for a metered credit pool, and remains a live grievance
  • Media-minute and AI credit meters are separate, so a busy month can produce bill shock
  • Avatar and video generation drain the entry tier's credits quickly
  • Their own docs warn MCP-driven editing burns credits faster than manual editing
  • No MP4 export through MCP, only publishing to a share link

Pricing: Free, then from $16/mo billed yearly (Freemium)

Rating: 4.5/5