How to Edit Video with Claude Without Setting Up ffmpeg, Skills, or Plugins

If you saw a tutorial where someone edits a video by getting Claude to run ffmpeg, here is the part the tutorial skipped: you don’t have to set any of that up to get the same result. Open chat.weftly.ai in a browser, upload your file, and pick what you want. The transcript, clips, or vertical shorts come back in minutes. No ffmpeg, no skills, no plugins, no tokens spent.

Those tutorials are genuinely good. They also show you the whole iceberg: installing ffmpeg, configuring Claude, loading skills and plugins, and then spending tokens every time the agent reads a file or fixes a broken command. That is fine if editing software is the hobby. It is a lot of overhead if you just want to ship one video a week and get back to creating.

Why We Built Weftly

Weftly came out of our own newsroom. We run The Woven Record, a grassroots news site, and we needed local journalists to be able to turn raw footage into video stories without first learning editing software. That is what chat.weftly.ai is. A reporter uploads what they shot, picks what they need, and gets it back, so the time goes into the story instead of the software.

We also edit our own videos and our Weftly demo videos, and for that we use the MCP server. We are comfortable working in Claude. We figured other people would like to use the same tools we do, but without building and maintaining them. We built both paths because we use both paths every week.

So before you decide to set all of that up, figure out which kind of creator you are.

Which Path Fits You

There are really three kinds of people who watch the “edit video with Claude” tutorials.

You want the result, not the setup. You don’t want to install anything or learn a tool. You want a transcript, a few good clips, or a vertical short, and you want to spend your time creating, not configuring. This is most solo creators, and it is the local journalists we built Weftly for. Use chat.weftly.ai. It is a web page. You upload, pick, and pay per job. There is nothing to maintain.

You already work in Claude and want it to do more without the overhead. You are comfortable with software and you like having the agent in the loop, but you don’t want to run ffmpeg locally and you don’t want editing to eat your token budget. Add the Weftly MCP server. Claude calls a tool, Weftly does the heavy processing on its own servers, and the finished file comes back. You skip the local install, and you stop paying in tokens for every ffmpeg retry. This is how we edit our own videos and our Weftly demos.

You want to configure everything yourself. You enjoy maintaining ffmpeg, wiring up skills and plugins, and spending tokens to drive the edit. That path works and you keep full control. If this is you, you probably already closed this tab, and that is fine. Weftly is not trying to replace it.

Weftly serves the first two. The rest of this post is for them.

What It Costs in Time

A sixty-minute episode turned into five vertical shorts plus a transcript takes about fifteen minutes (post upload) on Weftly, and almost none of it is your attention. You upload once, pick the clips, and the processing runs while you do something else.

The DIY route spends that same time differently. Installing ffmpeg, getting Claude configured, loading the right skills, and then waiting through the retries when a command does not parse the first time. The output can be just as good. The difference is how much of your day it takes and whether you are babysitting it.

What It Costs in Money

Pay per job, no subscription, the same prices whether you use the web page or the MCP server:

• Transcript: $0.50 audio, $1.00 video
• Summary: $0.75 audio, $1.25 video
• Find Clips: $2.00, returns ranked candidates with timestamps
• Each clip cut, horizontal or 9:16 vertical: $0.50

So that sixty-minute episode into five shorts is $2.00 + (5 × $0.50) = $4.50, start to finish. For how that compares to Descript, Castmagic, and the rest, see the 2026 podcast post-production cost comparison.

The DIY route looks free until you count the tokens. When Claude runs ffmpeg itself, every file read, command write, output check, and retry is tokens out of your budget. A flat $0.50 per cut is easier to predict than a token bill that grows every time a filtergraph fails.

Why Hosted Uses Fewer Tokens

This is the part the tutorials rarely mention. Getting Claude to edit video directly means the model is doing the work: reading the file, composing the ffmpeg command, inspecting what came out, and trying again when something breaks. Each of those steps consumes context and tokens. Not a problem if that’s the kind of thing you enjoy, but another piece of software to learn and manage if it’s not something you enjoy.

With Weftly, the processing happens on Weftly’s servers. The agent makes one tool call and gets the finished result back. The model is not parsing video or retrying commands, so the same edit costs a fraction of the tokens. That is the whole point of the MCP path: keep Claude in the loop without making Claude do the heavy lifting.

How to Start

If you want the simplest path, go to chat.weftly.ai, upload a file, and pick what you want. Pay by card per job. Nothing to install.

If you want Claude to do it, add the MCP server at api.weftly.ai/mcp to your host and ask in plain English. The MCP setup guide walks through it.

Either way, the setup the tutorials show you is optional. Skipping it gets you the same transcript, clips, and shorts, with less of your time and a bill you can predict.

---

Try it: chat.weftly.ai · Pay per job · No subscription · No setup · weftly.ai