Clip and schedule video from an AI agent
ClipAI ships an MCP server, so the whole pipeline is callable by an agent. Claude, or any MCP-compatible client, can import a long video, read the AI-detected clips and their scores, render one to a captioned 9:16 short, and schedule it to a connected account - from a single message. Seven tools, one API key, no clone or build step.
Install
Add this to your MCP client's configuration. It needs Node 20 or newer. Generate the key in ClipAI under Settings, then API keys.
{
"mcpServers": {
"clipai": {
"command": "npx",
"args": ["-y", "clipai-mcp"],
"env": { "CLIPAI_API_KEY": "clip_xxxxxxxxxxxxxxxxxxxx" }
}
}
}CLIPAI_BASE_URL is optional and defaults to https://clipai.online. It must be an origin - scheme and host, no path - and it must use https, with localhost allowed for local development. The server checks this at startup and exits with an explanatory error rather than running misconfigured.
The tools
import_youtube
Import a YouTube URL and start AI clip detection. Returns a video_id and a job_id.
list_videos
List the org's videos with their status and clip counts.
list_clips
List clips, optionally for one video: id, title, status, score, start and end.
get_clip
Fetch a single clip. This is what you poll after rendering.
analyze_clip_url
Explain why a public YouTube or TikTok clip works: score, band, hook type, reasoning and the rubric breakdown. One scoring run.
find_moments
Find more moments about a topic or pattern in a finished video (one extra scoring pass). Use after analyze_clip_url to find your own version of a clip that worked.
render_clip
Render a clip to a 9:16 short with burned-in captions. Pass reframe_focus: "track" to pan the crop with the speaker. To dub, pass dub_language and dub_consent: true, the user's confirmation that the dubbed speakers consented (clipai-mcp 0.2.1 and later). Returns a jobId.
list_social_accounts
List the connected social accounts available to schedule to.
schedule_clip
Schedule a rendered clip, or publish it now, to one of those accounts.
A typical agent run
- 01import_youtube
- 02list_clips (poll until clips appear)
- 03find_moments (optional: one more pass about a hook or topic, for example one analyze_clip_url explained)
- 04render_clip
- 05get_clip (poll until status is READY)
- 06list_social_accounts
- 07schedule_clip
Questions
What is the ClipAI MCP server?
It is an MCP server that exposes ClipAI's clipping pipeline as agent-callable tools, so Claude or any MCP-compatible client can import a video, list the AI-detected clips, render one, and schedule it to a connected social account from a single message. It is published to npm as clipai-mcp and needs Node 20 or newer.
How do I install it?
You do not need to clone or build anything. Point your MCP client at npx -y clipai-mcp and set CLIPAI_API_KEY in its environment. The configuration block on this page is the whole setup.
How does it authenticate?
With a per-org API key that you generate in ClipAI under Settings then API keys. It is sent as an Authorization bearer token. The key starts with clip_, carries the org owner's authority, and can be revoked at any time from the same screen.
Can I point it at a different host?
Yes. Set CLIPAI_BASE_URL, which defaults to https://clipai.online. It must be an origin - scheme and host with no path - and it must use https, with localhost allowed for local development. The server validates it at startup and exits with an explanatory error rather than running misconfigured.
Does the API work without MCP?
Yes. The same API key works for any programmatic call, not just this server. The MCP server is a wrapper over the ClipAI API, so anything an agent can do here you can also do directly over HTTP.
What does it cost?
MCP and API access are included in every ClipAI plan (from $12/month; plans differ only in videos a month), and you can cancel anytime. There is no separate agent or developer tier.
Related
What ClipAI does with your footage
One long video in. Scored clips out.
ClipAI watches the whole recording, scores every moment 0 to 100 and hands back captioned 9:16 clips, best first, ready to schedule.
