How to automate clipping with an AI agent: API, MCP and what to delegate
You can ask Claude or ChatGPT to clip, render and schedule your videos. The path is MCP, the protocol that plugs an assistant into a clipping tool. Here is what works, who offers it, what it costs, and where I would not let the agent decide on its own.
How to automate clipping with an AI agent: API, MCP and what to delegate
You can ask Claude or ChatGPT to analyze a video, clip it, render the best moments and schedule them, without opening the tool's dashboard. The path is MCP (Model Context Protocol), and in 2026 at least three well-known clipping tools offer an MCP server, Cut.Pro among them (on the Plus and Pro plans). The agent is great at the repetitive part of the workflow. Choosing what goes live is still a job I would keep for you.
MCP in one sentence
It is an open standard Anthropic released on November 25, 2024 to connect AI assistants to external tools. In practice, each function of the clipping tool becomes an action the assistant knows how to call, with a name, parameters and a description. You speak plain English, and the model decides the sequence of calls.
The difference from traditional automation (Zapier, a script) is that you do not write the sequence. You describe the outcome.
Who offers MCP for clipping today
Checked on September 26, 2026:
| Tool | How it connects | Plan required |
|---|---|---|
| Cut.Pro | Local through npx (Claude Code, Cursor, VS Code, Claude Desktop) or remote at mcp.cut.pro with OAuth (ChatGPT, Claude.ai) |
Plus and Pro |
| OpusClip | Official Claude connector, launched September 11, 2026, no API key needed | MCP connector listed on Pro and Business |
| Reap | Claude Desktop, Claude Code, Cursor, VS Code, Gemini CLI and others | "Every paid plan includes MCP, API, and CLI access" |
Sources: OpusClip changelog, OpusClip pricing, Reap MCP.
Being straight about it: Cut.Pro is not the cheapest way in here. Reap opens MCP on every paid plan, starting at US$ 9.99 a month billed annually, and OpusClip's Claude connector skips the key entirely. What our server does differently is coverage: it exposes all 44 v1 API endpoints, one tool per endpoint, from analysis to publishing. If the API can do it, the agent can.
What the agent can do on Cut.Pro
Per the public Cut.Pro MCP documentation, the tools cover:
- Account: workspace, plan, credit balance and ledger.
- Videos: list the library, upload your own file, delete.
- Clipping:
analyze_video(metadata and credit cost preview, free),submit_clipping(starts the job and charges), polling until it finishes. - Clips and templates: list clips sorted by score, apply a template in bulk.
- Renders: render to MP4, poll, get the download link, bundle several downloads.
- Transcription: transcribe with speaker labels and download as SRT, VTT or TXT.
- Publishing: create immediate or scheduled posts on connected accounts, publish now, retry a failed target.
Each tool is annotated as read-only, write or destructive, so the client knows when to ask for confirmation.
How to connect
ChatGPT or Claude.ai: add a custom connector pointing to https://mcp.cut.pro. The client opens a consent page, you paste your API key, done.
Claude Code:
claude mcp add cutpro --env CUTPRO_API_KEY=<your-key> -- npx -y @cutpro/mcp
Cursor, VS Code or Claude Desktop: the same JSON config with npx -y @cutpro/mcp and the key under env. Keys are created in your account settings, under API keys.
A real request
"Analyze this YouTube episode, tell me what it will cost, and if it is under 90 credits, clip it and render the three highest-scoring clips."
The model chains it on its own: analyze_video, applies your rule, submit_clipping, get_submission until done, list_clips, render_clip for the three, get_render and get_render_download. You get three MP4 links.
The phrase that saves the most money in that request is "tell me what it will cost". It turns the first call into a free lookup.
What is worth delegating
- The weekly batch. Every Monday, the new podcast episodes: analyze, clip, apply the channel template, render.
- Bulk templates. Thirty clips in the same caption style without opening each one.
- Transcripts for other uses. The episode SRT for YouTube, or raw material for a blog post.
- Scheduling. "Schedule today's five renders on TikTok and Instagram, one a day at 7 pm." The agent creates posts on accounts you already connected.
- Spend reports. "How many credits did I use this month, and on what?" comes straight from the ledger.
What I would not delegate
The final call on what goes live. The 0 to 10 score is a good filter and the model sorts well. But you are the one who knows a clip will upset a sponsor, or that the guest asked for something not to be published. An agent that publishes without review will, sooner or later, post the wrong thing to the right profile.
Spending without a ceiling. Starting a clipping job costs credits. Give the agent an explicit rule: always check cost first and ask above a threshold.
Keeping the key anywhere. The API key opens the whole workspace. It lives in an environment variable, never pasted into a shared doc, and if it leaks you revoke it in the dashboard.
Today's limits
- Connecting a new social account does not go through the API: each account is authorized inside the app. The agent publishes to accounts already connected.
- Publishing to YouTube does not exist on Cut.Pro. YouTube is a video source, not a destination. Publishing covers TikTok, Instagram, Facebook, X, LinkedIn, Bluesky, Pinterest and others.
- Live stream clipping is not among the MCP tools. Live monitoring stays in the dashboard, as covered in clipping live streams in real time.
- Per-key rate limits. The API returns what is left on every response; a well-configured agent honors
Retry-Afterwhen it hits the limit.
Is it worth it?
If you clip one video a week, no. The dashboard is faster than explaining to a model what you want.
For an agency with ten clients, for anyone who already lives in Claude Code or Cursor, or for anyone plugging clipping into a bigger flow (research, script, publish, report), MCP takes the work nobody likes off the table. It is the same logic as why the workflow became the product: when tools converge, the winner is whatever fits best into the rest of your day. If you run volume, a clipping agency shipping 100 videos a month shows where the real bottleneck sits.
The honest test: take the most tedious task of your clipping week and write, in one sentence, the result you want at the end. If the sentence fits, the agent can handle it.
Sources: Anthropic, introducing the Model Context Protocol · Cut.Pro MCP docs · OpusClip changelog · OpusClip pricing · Reap MCP
