Creator OS
claude codeclipper apicreator osmcpvideo clipping

Clipper API: Clip a Stream From Claude Code in 2026

The Clipper API (/v1/clipper) and clipper_ MCP tools let Claude Code clip a video link, score it, and post. Clippy is in private beta; here is the workflow.

Creator OS · November 18, 2026 · 9 min read

The clipper API is the part of Creator OS that turns a long stream into short vertical clips from Claude Code. Clippy, the Creator OS clipper, is reachable through the REST API at /v1/clipper and through clipper_ tools on the hosted MCP server. It is in private beta, so not every account has it yet.

stream link or file
        |
        v
  Claude Code  --(MCP: clipper_ tools)-->  Creator OS
        |                                      |
        |                               transcript read
        |                               moments cut
        |                               captions proofread
        |                                      |
        v                                      v
  clip list + Virality score  <------  rendered 9:16 clips
        |
        v
  post to Instagram, TikTok, YouTube Shorts
Clippy sits behind the API and the MCP server, so Claude Code drives the whole clip to post loop.

What the clipper API actually is (and why it lives in Claude Code)

Clippy is a hosted service. The clipping happens on Creator OS servers, and the API is how your own code or your AI app talks to it.

That design matters. A local script that tries to cut clips needs ffmpeg, a GPU, and a transcript pipeline on your laptop. Clippy runs the transcript read, the moment selection, the caption proofread and the render on the server. Claude Code is just the control surface. You describe the job in plain words, Claude calls the tool, and the clips come back with scores attached.

Clippy takes three kinds of source: a creator you find on Twitch, Kick or YouTube, a video link you paste, or a file you upload. It returns up to 10 clips per video. On a stream it scans up to 3 hours and skips a 4 minute intro on longer streams, which is a useful default when you point it at a full broadcast.

The same Clippy engine has other surfaces: the Agents hub in the web app, /v1/clipper in the REST API, and the clipper_ tools on the MCP server. Learn the workflow once and it transfers between them.

Get started for $0

Is Clippy available? Private beta, stated plainly

Clippy is in private beta and not yet open to every account. Saying it any other way would be misleading.

What that means in practice:

  • If your account does not have Clippy, the clipper_ tools and /v1/clipper will not be there for you.
  • There is no public queue list or waitlist page I can point you to. The docs at https://www.creatoros.ca/docs/mcp and https://www.creatoros.ca/docs/agents are where the current state is described.
  • The rest of Creator OS works without it. Posts, analytics, comments, DMs and the other agents are on every plan.

Two things are also worth stating up front. Clippy uses Claude on the server side to read transcripts and pick moments, so what you send it are links or files, not credentials. And you need the right to post clips of someone else’s content. Follow each creator’s clipping rules and each platform’s policies.

How to clip with Claude: the Clippy workflow from link to post

The loop is four steps: give Claude a source, let Clippy pick and cut, read the scores, post the keepers.

  1. Point it at a source. Paste a video link, name a creator on Twitch, Kick or YouTube, or upload a file. Ask for a number of clips up to 10.
  2. Let Clippy choose. Claude reads the transcript and scouts about three times as many candidate moments as you asked for. An editor pass then cuts each candidate on exact words, so cuts never start or end mid-word, and up to 3 dull stretches can be cut out of the middle. Each cut is scored and the best are kept.
  3. Read the Virality score. Every clip comes back with a 0 to 100 score and a report of what to fix. More on that below.
  4. Post or refine. Rate a clip Good clip or Not for my page, or say in plain words what to change. Clippy plans the edit and re-cuts just that part.

Two layouts change the output shape. The gaming layout finds the streamer’s facecam once per stream and renders a split screen: gameplay on top, facecam below, captions on the seam. The podcast layout follows whoever is talking from mouth movement, switches to a split screen when two people trade lines, and alternates a punched-in and a wide shot every few seconds on single-speaker shots.

Captions are word-by-word and animated. They are transcribed, then proofread against the source to fix misheard names, places and slang. There are 8 caption design presets and 12 fonts, with colours, position, words per line and a pop or box highlight. Audio is normalised to -14 LUFS and silence cutting is optional.

What the clipper_ MCP tools do inside Claude Code

The MCP server exposes Clippy as clipper_ tools, which means Claude Code can run the whole job without you touching a request body.

Connect it once. In Claude Code:

claude mcp add --transport http creatoros https://mcp.creatoros.ca/mcp --header "Authorization: Bearer cos_live_..."

Or install the plugin, which adds the MCP server plus skills, a live pane and a publish guard:

/plugin marketplace add kevinbadi/creatoros-claude-plugin
/plugin install creatoros@creatoros
/plugin configure creatoros@creatoros

After that, a prompt does the work. Something like: “Find the last stream from this creator, clip the 6 strongest moments for TikTok, use the podcast layout, and show me the scores before anything goes out.”

Keep the workspace boundary in mind. Each connection is scoped to one workspace, and on agency plans you connect one client per connection. An agent holding one workspace’s key cannot post to another. If you choose a read only connection at setup, the tools that read clips will work and the API refuses writes from it.

If you want to see MCP and Claude Code talking to real accounts end to end, watch the walkthroughs on the KevBuildsApps YouTube channel, and the launch video for the bigger picture.

Calling /v1/clipper directly from your own code

If you have your own pipeline, /v1/clipper is the REST surface for Clippy, and it uses the same API key and error shape as the rest of Creator OS.

Authenticate with a cos_live_ key pinned to one workspace. Errors come back as { error: { code, message, status } }, which makes retries and logging straightforward. The docs at https://www.creatoros.ca/docs carry the base URL and the current request shape.

A realistic use: a nightly job that watches a list of creators. When a new broadcast appears, your script calls Clippy, gets back a set of scored clips, filters for anything at or above your bar, and hands the winners to POST /v1/posts with content, platforms, media and schedule_at. Every render arrives with a caption, so the post body is mostly the media id and the copy Clippy wrote.

I am not going to invent request fields here. The endpoint exists, it is called /v1/clipper, and the docs describe what it accepts. Build against those.

Reviewing clip scores before you post

The Virality score is the reason to route clips through Claude instead of blind posting. It runs from 0 to 100 on a rubric, with a verdict attached.

Signal Weight
Hook 25
Retention 20
Topic 15
Emotion 15
Clarity 10
Shareability 10
Platform fit 5

The verdict is simple: post at 75 and above, fix first in the middle, skip under 45. The report tells you what to fix, and it is grounded in evidence rather than vibes. It looks at frames from the first 3 seconds and across the video, the word-by-word transcript, and measured stats.

A prompt that uses this well: “Clip this stream for Shorts, then for anything scoring 60 to 74 apply the report’s fixes once, re-render, and keep whichever version scores higher.” That is the auto polish loop, and it is why asking for the scores before publishing is worth the extra step. Clips you rate Good clip or Not for my page feed back into what Clippy picks next, once 3 or more clips have either a rating or 24 hour view numbers.

How this compares to CapCut MCP and YouTube MCP in Claude Code

These are different tools with different jobs. Describing them neutrally is the honest way to frame the choice.

An editing MCP is a bridge between Claude and a desktop video editor, so edits happen on your machine inside that editor’s timeline. A YouTube MCP typically exposes channel and video data from YouTube’s own APIs, which is useful for reading uploads and stats. Creator OS ships a hosted MCP server with clipper_ tools, a REST endpoint at /v1/clipper, and a publishing layer, so an agent can go from a stream link to scored vertical clips to posts on connected accounts without you opening an editor.

What you want Where it fits
Frame-level edits in a desktop editor An editing MCP driving that editor
Channel and video data from YouTube A YouTube MCP
Long stream to scored vertical clips to scheduled posts Creator OS Clippy via /v1/clipper or clipper_ tools
Cutting a raw talking-head clip into an animated short Open source skills at open-edits, rendered locally

For a fuller tour of an agent that repurposes long content, see building a content repurposing agent with Claude Code, and for the general agent setup read how to build a social media agent with Claude Code.

FAQ

How do I clip a stream with Claude Code?

Connect the Creator OS MCP server to Claude Code, then ask Claude to clip a stream from a link, a creator on Twitch, Kick or YouTube, or a file. Claude calls the clipper_ tools, Clippy transcribes, cuts on exact words and returns scored vertical clips. Clippy is in private beta, so the tools appear only on accounts that have it.

What is the clipper API endpoint?

Clippy is exposed at /v1/clipper on the Creator OS REST API, authenticated with a cos_live_ key pinned to one workspace. The docs at https://www.creatoros.ca/docs carry the base URL and the current request shape. I am not listing fields here because they belong in the docs, not in a guess.

Can you use CapCut MCP or YouTube MCP in Claude Code instead?

You can connect whichever MCP servers fit your workflow, and an editing MCP or a YouTube MCP does different work than Clippy. Creator OS provides a hosted MCP server with clipper_ tools plus publishing, so an agent can take a stream to scored clips to scheduled posts in one chain. Choosing between them depends on whether you want timeline edits, channel data, or a clip to post pipeline.

How do I get Clippy beta access on Creator OS?

Clippy is in private beta and not yet open to every account, so I cannot promise access or a timeline. Check https://www.creatoros.ca/docs/agents and https://www.creatoros.ca/docs/mcp for the current state. Everything else in Creator OS, including the API, the MCP server and the CLI, is included on every plan.

Do the clipper_ MCP tools work outside Claude Code?

Yes. The same hosted MCP server at https://mcp.creatoros.ca/mcp works with Claude, ChatGPT, Cursor, Windsurf and VS Code, with read only or read & write connections. The clipper_ tools are on beta accounts regardless of which AI app you connect from. Your own code can call /v1/clipper directly instead.

Get started

If you want the posting, scheduling, analytics and inbox side running while you wait on Clippy, start here: https://www.creatoros.ca/sign-up. Creator OS plans start at $14.99/month for 11 platforms plus Skool, with the API, the MCP server and the CLI included.

To see the agent side in action first, watch the walkthroughs on the KevBuildsApps YouTube channel and the launch video.

Get started for $0
Free trial for new subscribers. Cancel anytime.

Keep reading