Podcast MCP Video Workflow Example
Here is a complete assistant-led session on real material: a 60-minute interview episode goes in as a URL, and four captioned vertical clips come out, with every tool call shown. Podcasts are the ideal MCP source because the value is buried in conversation, exactly what moment detection is built to find.
Workflow
Submit the episode by URL
You paste the episode link and the brief ("4 clips for Reels, audience is indie hackers, skip the ad reads"). The assistant calls submit_video with the URL; for hosts that are not directly supported per list_supported_providers, it asks for the raw video URL or uses create_upload for a local file. Resubmitting the same URL within 24 hours reuses the existing run.
OpenClip processes while the assistant polls
For an hour of conversation, expect several minutes of processing: transcription, viral-moment detection, and speaker tracking for the video feed. The assistant polls get_video_status every 10 to 15 seconds and reports transitions (downloading, processing, completed) so you can go do something else.
Review the detected moments
list_clips returns the scored moments: for an interview episode that typically means the contrarian take, the origin story beat, the tactical breakdown, and the quotable exchange. Titles, hooks, and quotes are generated from the audio and marked as untrusted content, so the assistant presents them for judgment. You pick four.
Verify a quote before rendering
For the moment you are unsure about, the assistant pulls get_transcript with a window around the timestamp at word level, so you can check the guest's exact phrasing (and that the answer is not missing its caveat) before spending a render on it.
Render with the show's caption style
The assistant calls render_clip for each approved moment with your preset (confirmed once via list_caption_presets), polls get_render_status, and returns four clip URLs: vertical crops that follow the active speaker, word-level captions in your style.
Repeat next week in one message
With the show's settings saved as a processing agent (create_agent: tracker, captions, logo), next episode is "same as last week" plus a URL. get_usage shows the credit spend in minutes per episode so you can budget a season.
Key Metrics
One 60-minute episode
Source
5 distinct tools
Tool calls
Brief + approvals
Your active time
Features
Interview-tuned detection
Moment detection surfaces the exchanges that work as standalone clips: strong claims, stories, and answers with a beginning and an end.
Speaker-aware framing
Vertical crops follow whoever is talking, so two-person episodes do not become a static wide shot with unreadable faces.
Quote verification built in
get_transcript at word level lets you check exact phrasing before rendering, which matters when a guest's nuance is the story.
Sound-off ready
Word-level captions in a consistent preset carry the conversation for the majority of viewers who scroll on mute.
Weekly cadence, one message
A saved processing agent turns episode clipping into a recurring one-liner instead of a recurring project.
Season-level budgeting
Credits are minutes of processing, so a weekly one-hour show maps cleanly onto plan allowances (150 minutes on Starter, 300 on Pro).
Frequently Asked Questions
Clip this week's episode from your assistant
Connect OpenClip, paste the episode URL, and approve your four best moments before the next recording.