Edit Your Videos From Claude Code: A Podcast to Shorts Workflow With the Sentarys MCP

TL;DR: with the Sentarys MCP connected, one prompt in Claude Code turns a podcast link into edited Shorts, and follow-up prompts adjust a clip by its exact words, export it in 1080p and dub it. You review links instead of timelines.

Why edit from a coding agent

If you already live in a terminal, switching to a video editor for every clip is the slow part. Claude Code can call tools, wait for jobs and read results, which is exactly how clip work flows: import, analyse, adjust, export. It also reads files on your disk, so it can upload a recording you just exported without a browser.

The setup, once

claude mcp add --transport http sentarys https://api.sentarys.com/v1/mcp \
  --header "Authorization: Bearer snt_live_YOUR_KEY"

Create the key in Settings, API and MCP. The full guide for every client is in how to connect the Sentarys MCP.

1. From link to clips

Clip this episode into Shorts with the podcast template: https://www.youtube.com/watch?v=... Tell me what it will cost first.

Claude reads the prices with get_account, then calls create_clips_project. The server imports the video and starts the analysis on its own, so Claude only polls get_project until the status is ready. Each clip comes back with a title, the hook, scores and a preview link.

2. Edit a clip by what is said

Clip 2 starts slow. Cut the first sentence, make the hook "He lost everything in one night", switch to karaoke captions and zoom in on "federal prison".

Claude calls get_clip, which returns every caption line with its time, so it can place the cut and the zoom on the exact words. Then one edit_clip call applies all the changes. Edits are free and saved to the clip, so they also show in the editor in the app.

3. Export and review

Export it in 1080p and give me the link.

render_clip with mode export queues the final file; a minute or two later get_clip returns an exportUrl you can open or download for 10 minutes. Claude can also save it next to your project with curl.

4. More versions

Dub it to Spanish and Portuguese, and make 5 variations with different hooks.

dub_clip creates one new clip per language, with translated captions and voices cast to the speakers. create_variations returns a batch you can export in one go. Each call reports the credits it used.

5. Files on your disk

Upload ~/recordings/episode-12.mp4 and clip it.

start_upload returns signed part URLs; Claude splits the file, sends each part with curl and calls complete_upload, which starts the analysis. Nothing goes through a browser.

What stays in your hands

• Taste. The agent proposes, you pick the clips worth posting.
• Spending. Paid tools report their cost, and the agent can check the balance before long jobs.
• Fine edits. For frame-level work, every clip has an editor link in the app.

FAQ

Does this work in Cursor or ChatGPT too?

Yes. The tools are the same in any MCP client; only the setup differs.

Can it make long videos?

Yes: create_long_edit turns footage links into a long edit or recap, and create_faceless_video writes, narrates and edits a faceless video from a topic.

What happens if a job fails?

The tool answers with a clear error and the credits of the failed job come back automatically.