Clip every moment your fans timestamp in the comments
“12:34 is the best part.” Your fans are already telling you what to clip — your agent scans the comments for timestamps, cuts every flagged moment, and hands them back ready to post.
Run it with your agent
- Open Settings → Connectors in your client.
- Find OpusClip, click Add.
- Sign in with your OpusClip account in the OAuth window.
- Run
claude mcp add --transport http opusclip <url> - Start Claude Code and run
/mcp - Approve the OAuth sign-in with your OpusClip account
- Open Settings → MCP → Add new server in Cursor
- Paste the OpusClip MCP URL (Streamable HTTP)
- Sign in with OAuth on first use
- Open Settings → Connectors → Create in ChatGPT
- Paste the OpusClip MCP server URL
- Authenticate with OAuth using your OpusClip account
- Add OpusClip to
.vscode/mcp.json(type: http) - Open the MCP view in VS Code
- Sign in with OAuth when prompted
https://api.opus.pro/api/mcpWhat this workflow does
Scroll the comments under any decent-sized video and someone has left a timestamp: “12:34 is the best part,” “the bit at 8:05 killed me.” Those comments are free editorial direction — fans marking the exact seconds they'd share — and almost nobody acts on them. This workflow does: the agent reads the comments, extracts every timestamp, and clips each flagged moment from the source video with captions and vertical reframing.
Why it works
Every timestamp comment is a viewer who was moved enough to stop watching and write down a time code. That's a stronger signal than any prediction, because it's not a guess about what will resonate — it's a record of what already did. Clips built from fan-flagged moments also come with built-in social proof: the caption writes itself (“you asked for this moment”), and the commenter who flagged it becomes your first engaged viewer when you reply with the clip.
What you need
- OpusClip MCP (api.opus.pro/api/mcp) — does the clipping, captions, and reframing (Pro plan required for tool calls)
- A YouTube MCP or the YouTube Data API — reads the comment threads on your videos; comments are public, so any agent with YouTube access can scan them
- Claude, Cursor, or any MCP-enabled agent to run both and connect the dots
How it works
- Point the agent at a video. Yours, with a healthy comment section.
- It scans every comment for time codes. Timestamps like 12:34, ranges like 8:05–8:40, and phrases such as “the part where…” that it can match against the transcript.
- Flagged moments get grouped and ranked. Five comments pointing at the same minute count as one moment with five votes — likes on those comments add weight.
- Each moment becomes a clip. The agent cuts a 30–60 second window around each flagged time code, captioned and reframed for vertical.
- You get clips with receipts. Each one comes back labeled with the comments that requested it, ready to post — or to drop as a reply to the fan who asked.
Try this prompt
Scan the comments on this video of mine: [YouTube URL]. Find every comment that mentions a timestamp or a specific moment, group ones that point at the same part, and rank the moments by how many comments and likes flagged them. Then clip a 30–60 second window around each of the top 5, with captions and vertical reframe, and tell me which comment flagged each clip.
Tips
- Reply to the original commenter with the clip — “you asked, we cut it” replies routinely outperform cold posts.
- Run it on your most-commented videos first; the signal quality scales with comment volume.
- Ask the agent to also match “the part where…” phrases against the transcript — plenty of fans flag moments without exact time codes.
