Control Video Editing from AI Assistants - OpenClip
MCP Use Case

Control Video Editing from AI Assistants

Whatever assistant you already use can operate OpenClip. Chat apps like Claude and ChatGPT, editors like Cursor, research tools like Perplexity, and autonomous agents like Hermes, OpenClaw, or NemoClaw all connect to the same server at https://openclip.app/mcp and drive the same clipping pipeline. This page is the umbrella workflow that works everywhere.

AI-assisted production
Pro

Scenario

A creator or content team wants video clipping to happen wherever they already work: briefing an assistant in chat, scripting a loop in an editor, or handing the task to an autonomous agent, without maintaining a separate video workflow for each surface.

Workflow

1

Connect your client

Every MCP client takes the same URL, https://openclip.app/mcp, with OAuth sign-in. Claude and Perplexity add it under Connectors, ChatGPT under Apps & Connectors with Developer mode (paid plans), Cursor via a three-line mcp.json entry.

2

Or hand it to an autonomous agent

For chat agents like Hermes, OpenClaw, or NemoClaw, send this prompt: "Connect to the OpenClip MCP server at https://openclip.app/mcp. It's an OAuth 2.1 server with dynamic client registration: register a client, start the authorization flow, and reply to me with the sign-in link so I can approve access in my browser. Once you're connected, you can turn my videos into viral clips." Header-only agents can use a personal access token with https://openclip.app/mcp/key instead.

3

Brief once, with everything

Give the assistant the source URL, target platform, audience, clip count, and caption style in one message. It calls submit_video (checking list_supported_providers if the source is unusual) and confirms the job id.

4

Let the assistant watch the job

OpenClip transcribes, detects viral moments, and tracks speakers while the assistant polls get_video_status every 10 to 15 seconds and narrates progress. You do nothing during this stage.

5

Approve moments, render clips

The assistant presents the scored moments from list_clips (titles and hooks are marked untrusted content, shown for your judgment), you pick, and it renders each approval with render_clip in your preset, polling get_render_status until the URLs arrive.

6

Standardize for the team

Save the working setup as a processing agent with create_agent so every assistant, and every teammate's assistant, produces identical output. Watch spend with get_usage; credits are minutes of processing.

Benefits

One connection pattern covers every assistant, so switching or mixing clients costs nothing
No integration engineering: the first clip ships the day you connect
The assistant carries the brief and the polling; you only make editorial calls
Untrusted content isolation keeps transcript text from steering your assistant
Team presets via create_agent keep output consistent across people and clients

Key Metrics

About 5 minutes per client

Connection time

One: https://openclip.app/mcp

Server URL to remember

28, from submit_video to generate_ugc

Tools your assistant gains

150 processing minutes per month (300 on Pro)

Starter plan credits

Features

Client-agnostic by design

The server does not care which assistant calls it. Claude, ChatGPT, Cursor, Perplexity, and custom agents get the same tools and the same results.

Plain-language operation

"Five TikTok clips from this webinar, skip the intro" is a complete instruction. The assistant translates it into the right tool sequence.

The whole pipeline as tools

Submission, monitoring, moment review, rendering, transcripts, presets, and usage: every step of clipping is a callable tool, so nothing forces you back into a browser mid-flow.

Team-consistent output

Processing agents created once with create_agent apply the same tracker, captions, and branding no matter whose assistant submits the episode.

Safe enough to delegate

OAuth per user, a scoped mcp:use grant, and untrusted content isolation mean handing the mechanical work to an agent does not mean handing over judgment.

Costs you can see

Credits are minutes of processing, queryable any time with get_usage, and duplicate submissions within 24 hours reuse the existing run.

Frequently Asked Questions

The one you already use. Chat apps are best for briefing and review, Cursor is best for scripted loops, Perplexity is best when finding the source is part of the job, and autonomous agents suit recurring hands-off runs. The server and tools are identical.

Agents that can run an OAuth flow use the same connection prompt as everything else. Header-only agents mint a personal access token in OpenClip and call the https://openclip.app/mcp/key endpoint; treat the token like a password.

Yes. Each client holds its own OAuth grant against the same team, so a video submitted from Perplexity is visible to Claude via list_videos, and credits draw from the same pool.

Approving which moments get rendered and reviewing clips before publishing. Everything mechanical (submitting, polling, rendering, transcript pulls) is safe to delegate to the assistant.

Bring video editing to your assistant

Connect once and every long video becomes a conversation away from captioned, publish-ready clips.

Related Pages