For editing real footage by chatting with an AI video editing agent, OpusClip is our pick: its AI Producer works inside ChatGPT or Claude, and its MCP server and CLI let other agents clip and edit video too. Descript is the best fit for editing by transcript with an assistant, Kapwing for a browser editor with a chatbot, CapCut for a ChatGPT-driven draft in a familiar editor, Runway for generating new footage from chat, Submagic for captions, and Captions for a chat-based editor in a creator app. Two entries are categories rather than brands: a general assistant plus a video MCP server, and a do-it-yourself FFmpeg setup. We ranked each on the job it does best, what it costs, and where its free tier stops.

Updated October 28, 2026. Independence note: we make OpusClip and rank it first for editing footage through a chat-based agent. Every fact about the other eight entries comes from that vendor's own pricing page, docs or site, fetched on October 28, 2026, and we did not run hands-on tests of competitor products.

New to the idea? Our explainer what an AI producer is defines the category, and can ChatGPT edit videos covers the ChatGPT route step by step.

What is an AI video editing agent?

An AI video agent is software that takes a plain-language request ("cut the filler, add captions, make it vertical"), plans the edit, runs it against real footage and returns a finished video or an editable project. The difference from a template or a one-click filter is that the agent decides what to do next and accepts follow-up instructions on the same project. Most agents today work in one of three ways: a chat inside the vendor's own app, a plugin or connector that runs inside ChatGPT or Claude, or an MCP server that any compatible assistant can call.

That last route is why the list mixes products and categories. A general assistant plans and talks; a video tool does the rendering. Each entry below says which of those parts it supplies.

Best AI video editing agents at a glance

• OpusClip: best for editing talking-head footage by chat (AI Producer) and for agent-callable clipping (MCP server and CLI), from $15 a month, free plan yes (60 credits a month, watermark, no editing).
• Descript: best for editing by transcript with the Underlord agent, from $16 a month billed annually, free plan yes.
• Claude or ChatGPT plus a video MCP server: best for working inside the assistant already open, priced as the assistant plan plus the video tool's plan, free options vary.
• Kapwing: best for a browser editor with an AI chatbot, Pro listed at $16 a month, free plan yes (watermark).
• CapCut: best for a ChatGPT-driven draft that opens in a full editor, Pro at $19.99 a month, free plan yes, plugin not yet in the US.
• Runway: best for generating new footage from chat, from $12 a month billed annually, free plan yes (125 one-time credits).
• Submagic: best for captions and short-form polish via API or MCP, from $19 a month, no free plan listed.
• Captions: best for a chat-based editor in a creator app, Max at $24.99 a month, free tier shown without AI credits.
• FFmpeg with an LLM (DIY): best for developers who want full control, software cost $0, build time is the cost.

1. OpusClip: Best for editing real footage by chat and for agent-callable clipping

OpusClip approaches the agent idea from two directions. AI Producer is the chat-first editor: it takes raw footage of one person talking to camera, plans the edit and builds captions, motion graphics, B-roll, music and sound effects, and it can cut filler words and pauses. The editing happens through chat in the ChatGPT plugin or the Claude connector, and AI Producer's docs also cover Codex, Claude Code and Cursor Desktop. The ChatGPT plugin runs in Codex, which comes with paid ChatGPT plans, so a paid ChatGPT plan is needed. Output is one MP4 at up to 1080p and 30 fps in 9:16, 16:9 or 1:1. Our own guide says transcription, captions, B-roll search, voice enhancement and music use no OpusClip credits in the plugin, while AI scene restyle does.

The second direction is the developer and agent surface. OpusClip's hosted MCP server at mcp.opus.pro/mcp signs in with OAuth, and the help center lists 37 tools (29 workflow tools plus 8 media tools) covering clipping, captions, editing, exports, scheduling and generation. A command-line interface ships inside the opus-skills plugin for Claude Code and Codex and uses an API key. Under both sits the classic product: clipping long video into short clips with a Virality Score from 0 to 99, animated captions in 20+ languages, AI Reframe, scheduling and Premiere Pro and DaVinci Resolve export on Pro.

Pricing from the pricing page on October 28, 2026: Free is 60 credits a month with 1080p clips, a watermark and no editing. Starter is $15 a month (monthly billing only) with 150 credits. Pro is $29 a month, or $14.50 a month billed annually, with 300 credits. Business is custom. One limitation: AI Producer is built for one person talking to camera, and multi-camera edits and screen recordings are not supported yet. MCP tool calls need Pro Beta, Max or Business access, and the CLI needs an API key from a plan with API access.

Choose it if the job is turning raw talking-head footage into a finished video from a chat, or if an agent needs to clip and edit video through MCP. Skip it if the footage is multi-camera or screen-recorded, or if the goal is generating video from nothing. OpusClip is not a replacement for a full non-linear editor. More on the product is on our OpusClip MCP page and the AI video editor page.

2. Descript: Best for editing by transcript with an AI agent

Descript pairs its transcript-based editor with Underlord, which its site calls an all-in-one video agent and an AI co-editor. Requests such as "edit this down to 2 minutes" or "hide my jumpcuts by adding zooms" run against a project, and the Underlord page frames the whole approach as vibe editing. Descript also has a video editing MCP: the connector lets Claude, ChatGPT or any MCP client run edits, pull transcripts and publish, and the MCP page says there is no separate charge beyond the AI credits the plan already includes.

Descript's pricing page lists Free at $0 with 60 media minutes a month, 100 one-time AI credits and 720p local export. Hobbyist is $16 a month billed annually ($24 monthly) with 10 media hours and 400 AI credits a month. Creator is $24 a month annually ($35 monthly) with 30 media hours, 800 AI credits and 4K export. Business is $50 annually ($65 monthly). Plans are per person.

The limitation is stated on Descript's own MCP page: media export is the big gap, since an assistant can publish a composition and return a share link and download URL but cannot hand over an MP4 on request, and the MCP cannot import from YouTube links. The page also warns that automated editing can burn through credits faster than editing by hand. Choose it if the work is interviews, podcasts or webinars where editing text is the fastest route. Skip it if the plan is a pure chat workflow that ends in a downloaded MP4.

3. Claude or ChatGPT plus a video MCP server: Best for working inside the assistant already open

This entry is a category, not a product. A general assistant such as Claude or ChatGPT is good at planning, scripts and instructions; it has no timeline of its own. Adding a video MCP server or plugin gives it hands. Several vendors in this list publish exactly that: OpusClip's MCP server, Descript's MCP, Runway's MCP, Submagic's MCP server and CapCut's ChatGPT plugin each connect an assistant to a video tool, and the right pick depends on whether the job is clipping, transcript editing, generation or captions.

Cost is the sum of two bills: the assistant plan, and the video vendor's plan or credits. Setup is usually a one-time sign-in; vendors such as Runway and Descript describe adding the connector in the assistant and signing in with the vendor account. The upside is a single place for scripting, planning and editing, and iterating with one-line follow-ups. The downside is that results depend on both pieces: the assistant's plan limits, and what the vendor's tools can actually do. Directory availability also differs by client, so check the vendor's own install page for the assistant in use.

Choose it if a team already lives in Claude or ChatGPT and wants video as one more tool there. Skip it if a visual timeline is essential to review every cut, or if the assistant's plan limits would be hit by editing volume.

4. Kapwing: Best for a browser editor with an AI chatbot

Kapwing's AI video editor page describes a browser workspace where uploaded footage can be edited by text prompt (trim, subtitles, music, enhancement), refined by chatting with an AI assistant, and then opened in the full studio with timeline controls. The same page lists a Kapwing MCP with a "Try with Claude or ChatGPT" option, plus tools such as Smart Cut, Trim with Transcript, a clip generator and dubbing in 40+ languages.

Kapwing's pricing page lists Free at $0 with 10 credits, unlimited exports with a watermark, exports up to 4 minutes at 720p, and a 30-minute monthly export limit. Pro is listed at $16 a month with 1,000 credits, no Kapwing watermark, 4K export, projects up to 2 hours and a Brand Kit with custom fonts. Business is $50 a month with 4,000 credits, voice clones and lip sync. The pricing page offers monthly and annual billing (annual is marked as a 33% saving), and the raw page does not say which view the $16 figure belongs to, so check the toggle before budgeting.

The limitation: the free tier is watermarked and short. Choose it if the team wants one browser tool covering subtitles, resizing, generation and a chatbot. Skip it if the goal is a pure chat workflow with no editor open, or if free exports must be watermark-free.

5. CapCut: Best for a ChatGPT-driven draft that opens in a full editor

CapCut's CapCut x ChatGPT page describes a plugin that turns a prompt or existing media into an editable draft inside ChatGPT, then lets edits continue in chat (shorten scenes, reorder clips, change captions) or in the CapCut editor itself. It draws on CapCut's template and effects library, and the page says the CapCut MCP connection carries supported requests into the editable project.

CapCut's pricing page lists Free at $0 with basic editing tools and free templates, Pro at USD 19.99 a month (1,200 monthly AI credits, premium templates, 1 TB storage), Ultra at USD 79 a month, and a Pro yearly price of USD 179.99. The page notes that displayed pricing is for reference only and final pricing appears after sign-in.

The limitations are on the plugin page itself: it is designed for the ChatGPT desktop app on PC, and availability is currently outside the United States with US availability described as coming soon. It also says free allowances should be checked in the product. Choose it if the team is outside the US, already edits in CapCut and wants ChatGPT to produce the first draft. Skip it if US availability today is a requirement, or if the footage is a talking-head video that needs an agent to finish the whole edit.

6. Runway: Best for generating new footage from chat

Runway sits at the generation end of the list. Its MCP page describes generating images and videos from inside ChatGPT, Claude, Cursor and other compatible agents, with models such as Gen-4.5, Seedance 2.5, Kling 3.0 and Veo 3.1 available based on the account plan, and generations billed from the same Runway credits as the app. Runway's pricing page also lists an agentic creative collaborator among the paid-plan features.

Pricing from Runway's pricing page: Free is $0 with 125 one-time credits. Standard is $12 a month billed annually ($15 monthly) with 625 credits a month and no watermarks, which the page translates to 52 seconds of Gen-4.5 video. Higher tiers exist; see the pricing page for current figures.

The limitation is scope. The MCP page is about creating new images and clips, not about cutting, captioning and finishing footage someone already shot, and credits are consumed per generation, so volume gets expensive quickly. Choose it if the job is concept video, B-roll or product shots made from a prompt. Skip it if the job is editing existing talking-head footage.

7. Submagic: Best for captions and short-form polish through an API or MCP

Submagic's docs describe an API that generates AI-powered captions and effects, and they include an MCP server at api.submagic.co/mcp that connects Submagic to Claude, Claude Code, Cursor and other MCP clients. The documented API covers creating caption projects, generating Magic Clips from a video, updating projects, exporting, and publishing or scheduling to YouTube, TikTok, Instagram, X, Facebook and LinkedIn. Claude connects with a one-URL connector; other clients pass an API key as a bearer token.

The pricing page lists Starter at $19 a month (45 credits, which it equates to 15 videos, with a 2-minute video duration), Pro at $39 a month (120 credits, 40 videos, 5-minute duration) and Business at $69 a month (300 credits, 100 videos, 30-minute duration, 4K and 60 fps). The comparison table shows 10 API minutes a month on Starter and Pro and 100 on Business. No free plan appears on the page.

The limitation is shorter source length on lower plans and no free tier. Choose it if captions and polished short-form styling are the main job and an agent should trigger them. Skip it if the footage runs longer than a plan's duration cap, or if an assistant should do the actual cutting and storytelling.

8. Captions: Best for a chat-based editor in a creator app

Captions lists a chat-based editor among the features of its Max plan, alongside AI Edit, long-video clipping, AI actors and generated B-roll, music and sound effects.

Captions' pricing page lists Max at $24.99 a month with 500 credits a month, and Frontier tiers at $69.99, $139.99 and $279.99 a month. The page says the prices shown reflect iOS plans only, and its comparison table lists no AI usage credits on the free tier.

The limitation is that the pricing page states its prices reflect iOS plans only, so other platforms need a separate check, and no MCP or plugin appeared on the pricing page we fetched (we did not search further). Choose it if a chat editor inside a creator app is the appeal. Skip it if an assistant like Claude or ChatGPT needs to drive the edit.

9. FFmpeg with an LLM (DIY): Best for developers who want full control

The do-it-yourself route pairs an LLM or coding agent with FFmpeg, the open-source command-line multimedia framework. The agent writes and runs commands for trims, concatenation, resizing, filtering and encoding. FFmpeg's own site describes it as able to decode, encode, transcode, mux, demux, stream, filter and play nearly any media format, and licenses it under the LGPL 2.1 or later, with optional parts under the GPL.

The software costs nothing. The real cost is build time: captions styling, speaker framing, B-roll sourcing, review UI and error handling all have to be built or assembled from other tools, and nothing ships with a preview. Quality and consistency depend on prompts and scripts that someone has to maintain.

Choose it if a developer is comfortable owning the pipeline and wants exact control, or if the job is a repeatable batch (trim, convert, normalize). Skip it if non-technical teammates need a visual review step, or if the goal is polished social video without writing code.

How we compared these tools

We compared on October 28, 2026 using each vendor's own pricing page, docs and product pages, fetched directly. The criteria were: whether an agent or chat is the main way to edit, which assistants it works in, what kind of footage it handles (existing footage or generated), price and free-tier limits, and one stated limitation per tool. We did not run hands-on tests of competitor products, and nothing here is a quality ranking of output. Prices change often and some pages offer both monthly and annual views, so each entry notes the billing basis where the page made it clear.

OpusClip is ranked first for the job named in the title: editing footage through a chat-based agent, with MCP and CLI access for other agents. Different jobs favor different entries, and the list below shows how.

How to choose

• Talking-head footage to a finished video, from a chat: OpusClip AI Producer.
• Interviews and podcasts edited as text: Descript with Underlord.
• Already working in Claude or ChatGPT: pick an MCP server or plugin that matches the job (see the next section).
• Browser editor with a chatbot and generation tools: Kapwing.
• Existing CapCut workflow, outside the US: CapCut x ChatGPT.
• New footage from a prompt: Runway.
• Captions through an API: Submagic.
• Creator-app users who want chat edits: Captions.
• Engineers who want control: FFmpeg with an LLM.

Best MCP server for video editing: which one fits which job

The best MCP server for video editing depends on what the agent should do with video. For clipping long recordings into short, scored, captioned clips and editing them, OpusClip's server is built for that job and also exposes scheduling and exports. For editing by transcript and composition, Descript's MCP is the match. For generating new clips and images, Runway's MCP is documented for exactly that. For captions and short-form styling, Submagic's MCP is. Directory listings and client support differ, so confirm in each vendor's docs that a given assistant is supported before committing. For the ChatGPT side specifically, see our roundup of the best ChatGPT plugins.

How to edit videos by chatting: a typical workflow

Most chat-based editors follow the same loop. First, connect: install the plugin or connector in the assistant and sign in to the video vendor once. Second, upload footage and write a short brief that names the platform, audience and mood. Third, review the first cut and ask for one change at a time. Fourth, open the preview or editor for final adjustments, then export. The tools differ in what they hand back (a finished MP4, an editable project, or a share link) and in whether footage is edited or generated, so check both before choosing.

Red flags to watch for

• Watermarks on free exports: Kapwing's free exports and OpusClip's free clips carry watermarks, so test with real output before committing.
• Credits that run out fast: automated and generative edits spend credits quickly, and Descript's own MCP page warns about this.
• Export gaps: an agent that edits but cannot return an MP4 on request leaves a manual step.
• Regional or client limits: CapCut's plugin page lists non-US availability; OpusClip's and CapCut's ChatGPT plugins run in the ChatGPT desktop app.
• No manual override: prefer tools that open an editor or preview so the last five percent can be fixed by hand.
• Generation passed off as editing: a tool that creates new footage is not editing the footage already shot; check which one is on offer.

Frequently asked questions

What is an AI video agent?

An AI video agent is software that takes a plain-language request, plans the edit and carries it out on a video, accepting follow-up instructions on the same project. Examples in this list include OpusClip's AI Producer and Descript's Underlord.

Can I edit videos by chatting?

Yes. Edit videos by chatting with OpusClip AI Producer in ChatGPT or Claude, with Descript's Underlord, with Kapwing's AI assistant or with Captions' chat-based editor. Each differs on footage type, export options and price, as described above.

What is the best AI video editing agent in 2026?

For editing talking-head footage through chat, we rank OpusClip's AI Producer first, with the caveat that we make it. For transcript-based editing, Descript is a strong alternative, and for generated footage, Runway.

Can ChatGPT edit videos?

Not by itself into a finished video. ChatGPT can plan an edit and write scripts and captions, but editing needs a plugin or connector such as OpusClip AI Producer, CapCut x ChatGPT or Runway, which each add different abilities. Our guide can ChatGPT edit videos has the steps.

What is the best MCP server for video editing?

It depends on the job: OpusClip's server for clipping and editing existing long video, Descript's for transcript-based editing, Runway's for generating clips, and Submagic's for captions. All four publish their own setup instructions.

Is there a free AI video editing agent?

Several have free tiers: OpusClip Free (60 credits a month, watermark, no editing), Descript Free, Kapwing Free (watermarked) and Runway Free (125 one-time credits). AI Producer's page says editing can start for free, but the ChatGPT plugin runs in Codex, so a paid ChatGPT plan is needed. Every free tier has limits listed on the vendor's pricing page.

Do I need to know how to code to use these?

No for AI Producer, Descript, Kapwing, CapCut, Runway and Captions, which work through chat or an app. MCP servers and CLIs such as OpusClip's and Submagic's API are aimed at developers and agent builders, though connector-based setups need only a sign-in.