# Compare AI video editing: MCP servers, APIs and editing apps

> cuttingfloor is editing tools for your agent, not an editor with an agent inside. As of November 2026, from public docs, checked 2026-11-01. "?" means not confirmed in the product's own docs.

Canonical page: https://cuttingfloor.ai/compare

## Two kinds of AI video editing

- An editor with an AI inside: Descript (Underlord), CapCut, VEED, Kapwing, OpusClip, invideo. You open their app and their AI works in it, on their credits. Agent access, when offered, is often a way to send a request to that AI.
- Editing tools for your agent: cuttingfloor. You stay in Claude, ChatGPT, Codex, Cursor or Kazi and connect one open-source MCP server, hosted or self-hosted. The agent makes each edit (cut, trim, caption, mix, effects, export) and sees the result after every edit.

Forwarding a prompt: Descript's API and MCP have one editing call, a job that takes a written request ("remove the section from 1:30 to 2:15") and hands it to Underlord. There is no cut, trim, caption or render call, and no picture of the result is documented coming back. Import, search, publish and transcript export are direct calls. VideoDB's call_director works the same way.

Why editing tools matter:

1. Control: each tool does exactly what its arguments say.
2. The edit stays editable: every edit is a new timeline version, earlier ones stay, and the edit exports as FCPXML, OTIO or EDL for Final Cut Pro, Premiere or Resolve.
3. Your agent’s judgment: the agent you picked, with your instructions and context, decides the cuts.
4. No lock-in: open source (AGPL-3.0); hosted and self-hosted are the same Docker image.

Seeing the result: Frapea (contact sheets), Tella (frame checks) and the Resolve and Premiere bridges (still exports) also let an agent look. cuttingfloor returns a preview with every edit without asking (frames, or a waveform with measured loudness), plus qa.check (gaps, black frames, silences, loudness, cuts through words) and media.seek (find a moment in plain words).

## Video servers and APIs an agent can call

| Product | Hosted or local | Open source | Editing tools or a prompt | Agent sees its result | Cut by words | Audio and voice | Keys or accounts you bring | Export to an editing app | You pay per |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| cuttingfloor | Both: hosted, or the same Docker image on your own machine | Yes, AGPL-3.0 | Editing tools: cut, trim, move, caption, mix, effects, export. Every edit is a new timeline you can go back to | Yes: a preview comes back with every edit (frames, or a waveform with loudness measured), plus qa.check for exact checks and media.seek to find a moment in plain words | Yes, cut by word timings | Gain and pan keyframes, fades, EQ, compressor, voice denoise, master limiter, stems, voiceover with timed words, lip-sync | Only for lip-sync (your sync. labs key). Self-hosted, the visual-judgment tools also need a TypeSafe key | FCPXML, OTIO and EDL, plus SRT and VTT captions | Render minute. Editing the timeline is free. Self-hosting is free |
| Descript MCP | Hosted (Descript sign-in) | No | A prompt: editing is a written request to Underlord. Import, search, publish and transcript export are direct calls | Not documented. Publishing returns MP4 links | Transcript export, yes. Cuts go through an Underlord prompt | Studio Sound, filler removal and voice translation, through Underlord | None | ? | Media hours and AI credits, two meters |
| invideo MCP | Hosted (sign-in) | No | Editing tools on a multitrack timeline, says invideo. The full tool list is not published | ? | ? | Mixing with buses and faders, and generation models (vendor claim) | None | ? | Credits |
| OpusClip MCP | Hosted | No | Jobs: submit a project to clip, caption and reframe, then edit one clip with a JSON edit script | ? | Word timings in the transcript. Edits stay inside one clip of one video | Bleeping (censor jobs) | None | Premiere XML | Credits (processing minutes); renders through the MCP per minute |
| Kapwing MCP | Hosted (Kapwing sign-in) | No | A few direct tools: add subtitles, translate subtitles, dubbing, upload and download. A call to its AI, Kai, is planned | No | No. Subtitles and transcripts only | Dubbing with voice cloning | None (Pro plan or higher) | ? | Seats, with AI features on workspace credits |
| Submagic MCP | Hosted, with an API key (no sign-in flow) | No | Project settings: templates, presets, media insertions and transcript fixes before a render. Not timeline edits | ? | Transcript fixes before render | ? | A Submagic API key | ? | Credits (about three a video) and API minutes |
| Frapea | Runs in your browser tab; the agent pairs with it through a code | ? | Editing tools: add, split, trim, move, ripple, effects, transitions, keyframes, multicam | Yes, on request: contact sheets of the timeline and media, a timeline check, loudness | Yes, word-level transcript and word removal | Volume keyframes, loudness, voiceover (in-browser voices or your ElevenLabs key), sound effects | Optional: ElevenLabs, Pexels, Freesound | Not documented | No credits, says Frapea |
| DaVinci Resolve MCP bridges | Local: Resolve open on the same machine (most need Resolve Studio) | The bridges, mostly MIT. Resolve itself is closed | Editing tools that mirror Resolve’s scripting API, hundreds in the largest | Partly: frame exports to a file, marker thumbnails | Yes in the largest bridge, with optional Whisper | Fairlight, voice isolation | None, but a Resolve Studio licence ($295, one time) | Edits your Resolve project directly | Free bridge, plus the Resolve Studio licence |
| Premiere and After Effects bridges | Local: the Adobe app open on the same machine | The bridges, MIT. The Adobe apps are closed | Editing tools for Premiere (hundreds) or After Effects comps, layers and keyframes | Partly: still-frame exports in Premiere. None documented for After Effects | Partly: transcript cleanup and caption tracks in Premiere | Premiere audio keyframes and ducking | None, but an Adobe licence | Edits your Premiere or After Effects project directly | Free bridge, plus Adobe Creative Cloud |
| FFmpeg MCP servers | Mostly local, a few hosted | Mostly MIT; Kinocut is Apache-2.0 | File operations: each one writes a new file. No timeline | Partly: some extract frames or storyboards | Mostly no; Kinocut makes Whisper subtitles | Extract and convert audio; Kinocut adds loudness, mastering and local text to speech | None locally; hosted ones need their own key | ? | Free locally |
| Shotstack | Hosted | No | Render: the agent writes the whole edit as JSON and gets a video back. Templates and generated assets | Not documented for the agent. A preview canvas is shown to you | Not documented | Generated audio assets | Your Shotstack account | ? | Render minute ($0.20 pay as you go, or $39 to $499 a month) |
| Creatomate | Hosted | No | Render: templates and render jobs | No (a free dry-run check only) | Not documented | Not documented in the MCP | Your Creatomate account | ? | Render credits |
| JSON2Video | Local command-line server that calls its cloud | ? | Render: send a JSON movie, check it, use templates | No | ? | ? | Your JSON2Video key | ? | Credits (600 free, about 10 minutes with a watermark) |
| HyperFrames (HeyGen) | Local command line and skills; a hosted MCP in beta | Yes, Apache-2.0 (the framework) | Code: the agent writes the video as HTML, previews it and renders it | Partly: a local preview loop | ? | ? | A HeyGen account for the hosted server | ? | Free framework; HeyGen credits when hosted |

## Editing apps, feature by feature

Marks: Y = Yes, N = No, P = Partly, $ = Higher plan or metered, ? = Not confirmed. "Higher plan or metered" means a paid tier or separate credits or minutes.

| App | Filler-word removal | Captions with word highlight | Reframe | Background removal | Voice cleanup | Voice and music split | AI voiceover | Lip-sync | Beat sync | B-roll | Agent access | Self-host |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| cuttingfloor | Y | Y | Y | Y | Y | Y | Y | Y | Y | P | The whole product is tools for your agent, over MCP | Yes, one Docker image |
| Descript | Y | Y | ? | Y | Y | ? | Y | ? | ? | Y | MCP and API: a prompt to Underlord | No |
| CapCut | ? | Y | ? | $ | ? | ? | Y | ? | Y | ? | No official MCP or editing API; community servers write draft files | No |
| VEED | Y | Y | ? | Y | Y | ? | Y | Y | ? | ? | An API of AI models (dubbing, lip-sync, clean audio); no editing tools | No |
| Kapwing | ? | Y | ? | Y | Y | ? | Y | Y | ? | Y | MCP: subtitles, translation and dubbing | No |
| OpusClip | $ | Y | Y | ? | $ | N | Y | ? | N | Y | MCP: clip jobs and an edit script for one clip | No |
| Submagic | $ | Y | ? | ? | Y | N | ? | N | ? | Y | MCP with an API key: project settings and templates | No |
| Captions | ? | Y | ? | ? | Y | ? | Y | Y | ? | Y | API for captions and avatars; no MCP found | No |
| Runway | N | N | N | ? | ? | ? | Y | ? | N | Y | MCP for generating and restyling; no timeline edits | No |
| Riverside | $ | ? | ? | ? | Y | N | ? | ? | N | Y | MCP in beta on higher plans: plain-language edits | No |
| Adobe Premiere | Y | P | Y | P | Y | ? | ? | ? | N | P | Adobe’s Claude connector (whole jobs, no timeline); community desktop bridges | No (a desktop app) |
| DaVinci Resolve | ? | $ | $ | $ | $ | $ | $ | N | $ | N | No official MCP; community bridges (outside scripting needs Studio) | No (a desktop app) |
| Canva | ? | ? | P | ? | ? | N | ? | N | ? | ? | MCP for designs; no video timeline tools found | No |
| Pictory | ? | Y | P | N | ? | N | Y | N | N | Y | Video API; an MCP announced in 2025, status not confirmed | No |
| invideo | ? | ? | ? | ? | ? | ? | Y | ? | ? | Y | MCP on the live multitrack timeline | No |

How cuttingfloor does each:

- Filler-word removal: Your agent cuts the ums and restarts by their word timings.
- Captions with word highlight: caption.add takes word timings and highlights each word as it is said.
- Reframe: Your agent reframes from face boxes it measures.
- Background removal: create.cutout, built in, no key.
- Voice cleanup: Voice denoise, EQ and compressor filters, with loudness measured.
- Voice and music split: create.stems splits voice, drums, bass and music (Demucs).
- AI voiceover: create.voiceover, with timed words (Kokoro), no key.
- Lip-sync: create.lipsync, on your own sync. labs key.
- Beat sync: media.analyze finds the beats; your agent cuts on them.
- B-roll: Partly: no stock library. Your agent places your footage or graphics it makes.

None of these apps asks for your own AI provider key; cuttingfloor does for lip-sync only (sync. labs). Motion graphics are left out of the table: they could not be confirmed across the apps. cuttingfloor's create.motion renders animated HTML with transparency.

## Pick it when

- cuttingfloor: An open-source MCP server of video editing tools for the agent you already use. Pick it when you want your agent to do the whole edit on your own footage, check its own work, and leave you an edit you can still change or open in your editing app.
- Descript MCP: Hosted access to Descript; editing goes to its built-in AI, Underlord. Pick it when you or your team already edit in Descript and want to hand Underlord a request from your agent, then finish by hand in Descript.
- invideo MCP: Hosted access to the invideo timeline you see live, with generation models. Pick it when you want generated footage from many models and want to watch the agent work in the same timeline you edit.
- OpusClip MCP: Hosted clip jobs: long video in, short clips out. Pick it when you want many short clips cut from a long video and scheduled to your socials, with the least effort.
- Kapwing MCP: Hosted subtitles, translation and dubbing for one video in Kapwing. Pick it when you need subtitles, translations or dubbing on a single video and already pay for Kapwing.
- Submagic MCP: Hosted captions and short-form templates. Pick it when you want styled captions on Shorts from a preset, fast.
- Frapea: A browser editor your agent pairs with over MCP. Pick it when you want an editor you can also work in by hand while your agent edits in the same tab.
- DaVinci Resolve MCP bridges: Community servers that drive Resolve on your computer. Pick it when you already edit in Resolve Studio and want your agent working in your real project, next to the best colour tools there are.
- Premiere and After Effects bridges: Community servers that drive Adobe’s desktop apps. Pick it when your projects live in Premiere or After Effects and stay there.
- FFmpeg MCP servers: Community servers that run FFmpeg commands on files. Pick it when you need one-off conversions, trims or audio extracts on your own computer, for free.
- Shotstack: A cloud render API with an MCP server in beta. Pick it when you already know the edit and need it rendered at volume from data.
- Creatomate: Templates designed in its editor, filled in and rendered through an API. Pick it when one branded template needs many variations.
- JSON2Video: A JSON-to-video API with a local command-line MCP. Pick it when simple videos are assembled from data.
- HyperFrames (HeyGen): An open-source framework that turns HTML into video, with agent skills. Pick it when you want motion graphics as HTML in an open-source framework you run yourself.
- Tella: A screen recorder with a hosted MCP of editing tools that checks frames of the edited video. Pick it when your videos are screen recordings.
- Palmier Pro: An open-source (GPL-3.0) macOS editor with a local MCP that renders frames to check. Generation is paid and closed. Pick it when you edit on a Mac and want a local app your agent and you share.
- Remotion: A React framework for making videos in code, with agent skills. Companies pay for a licence. Pick it when you want the video to be a codebase you keep and maintain.