cuttingfloorStart free

cuttingfloorthe video editing agent for Claude, Cursor and ChatGPT

Video editing,
solved.

Your agent edits video now. Connect Kazi, Claude, Cursor or ChatGPT to one MCP server and say what you want cut. It imports your footage, reads the transcript, trims, captions, mixes and hands back a finished file.

Start free for 3 daysWatch it edit
Works in
  • Kazi
  • Claude
  • Claude Code
  • Cursor
  • ChatGPT
  • Codex

Bring the agent you already use. Sign in, then add one server address; the setup for each is below.

01 / The floorAgentic video editing: your agent, one MCP server, every edit a call
02 / A session

Agentic video editing, one tool call at a time. Here is one.

  1. You

    Cut my 40-minute interview into a 60-second vertical teaser. Captions on, loud enough for phones.

  2. Agent

    On it. Importing the footage and reading the transcript first.

  3. media.import { source: { url: "…/interview.mp4" } }
    m_7f2c · 40:12 · 1080p · ready
  4. media.analyze { kind: "transcript" }
    6,412 words, timed
  5. Agent

    I have the transcript. Three passages hold the argument, so I’m taking those.

  6. timeline.create { 1080x1920, 30fps }
    tl_01 · empty
  7. clip.add { words: [812, 861] }
    V1 00:00–00:14
  8. clip.add { words: [2209, 2260] }
    V1 00:14–00:31
  9. clip.add { words: [5120, 5198] }
    V1 00:31–00:58
  10. Agent

    Checking my own work before I hand it over.

  11. caption.add { in_clip: { clip_id, segments } }
    84 captions on V2
  12. audio.set_master { gain_db: 6, limit_db: -1 }
    measured −14.1 LUFS
  13. qa.check { timeline_id: "tl_06" }
    no gaps, no black frames, sync ok
  14. export.video { preset: "social_9x16_h264" }
    teaser.mp4 · 00:00:58:12 · signed link
  15. Agent

    Done. 58 seconds, captioned, −14 LUFS, and it passes its own checks. Here’s the file.

Every edit is a tool call. Every call returns a timeline and a picture of it, so the agent sees what it made before it ships.

03 / In your agent

Plug cuttingfloor into the agent you already use. It becomes your editor.

Take 1: waiting to start Take 1: the bounce Take 1: winding down Take 2: getting ready Take 2: arm rolls Take 2: shimmy Take 2: settling back Take 3: counting in Take 3: the spin Take 3: the drop and the step-out Take 3: holding the stance Take 4: dancing in to the camera Take 4: hit the pose Take 4: [laughs] “Did I get it?” Room sound under every move

motion-party-short.mp4 · 0:23
Kazicuttingfloor

  1. Make these four takes into one dance video, cut to a beat. Don't use a real song, I'll get copyright struck.

    I’ll make you an original beat so there’s nothing to get struck, then cut your four takes to land on it, with zoom-ins, a slow-mo into the big move and a hold on your final pose.

    • Used cuttingfloor › media.import four takes
    • Used cuttingfloor › render.frames looked at every take
    • Used cuttingfloor › code.run made an original beat
    • Used cuttingfloor › media.analyze found the beats
    • Used cuttingfloor › edit.batch a cut on every bar, 7 shots
    • Used cuttingfloor › clip.set_speed slow-mo into the step-out
    • Used cuttingfloor › edit.batch zoom-ins, pose freeze
    • Used cuttingfloor › audio.set_master louder, peaks under −1 dB
    • Used cuttingfloor › render.frames checked every cut
    • Used cuttingfloor › export.video vertical, 0:23

    Your 23-second cut is ready: an original 104 BPM beat, a cut on every bar, a slow-mo into the step-out and a freeze on your final pose. Want the move names on screen?

  2. Put the move names on screen and make the ending pop.

    I’ll put each move’s name up as you hit it, big enough for Shorts, and turn your final pose into a flash.

    • Used cuttingfloor › create.motion animated end flash
    • Used cuttingfloor › edit.batch 6 move names and the flash
    • Used cuttingfloor › text.set bigger sign-off
    • Used cuttingfloor › render.frames checked the titles
    • Used cuttingfloor › export.video vertical, 0:23

    Done: every move’s name pops up as you do it, and the pose hits with a flash and a “MOTION PARTY” stamp. Want a loop-friendly ending too?

Shown in your agent: Claude, ChatGPT or Kazi. cuttingfloor does the editing and hands back the file; your agent shows it to you.

Watch the before and after · All demos

04 / Ask for

Say it like you’d say it to an editor.

No prompt craft, no timeline. Rescue a take, cut it down, or make what you don’t have.

05 / What it does

62 tools.
No timeline UI.A video editing API built for agents: granular primitives it composes, plus code when it needs a loop.

  1. Cut by what was said.

    Transcribe once, then cut on words instead of timecodes. The agent reads the transcript and lifts the lines it wants.

  2. The review loop

    It looks before it ships.

    Stills and filmstrips after every edit, a review render it can watch, and exact checks for gaps, black frames, clipping and sync. Other video agents apply an edit and hope; this one sees what it made and fixes it first.

  3. Captions that keep time.

    Captions come from the transcript and follow every speed change and cut. Karaoke, word pop, typewriter, boxes, outlines.

  4. Mix like an engineer.

    Ride gain with keyframes, crossfade at the cut, and set the gain and limiter on the whole mix, with the loudness measured after every change.

  5. Make what you don’t have.

    Animated titles and graphics from HTML, stills, mattes, a voiceover with timed words, a subject without its background.

  6. Code when tools aren’t enough.

    For loops and arithmetic over the timeline, the agent writes one script. Inside it, every tool is a function.

tools/call01/06
media.analyze { media_id: "m_7f2c…", kind: "transcript" }

clip.add { timeline_id: "tl_01…", track_id: "V1",
  media_id: "m_7f2c…", source: { words: [812, 861] } }

// whisper.cpp, on our servers
06 / Start

Three steps.

  1. 01

    Connect your agent.

    Add cuttingfloorai.apifunnel.ai/mcp as an MCP server in Claude, Cursor, ChatGPT or Kazi. Your client opens a sign-in page; Google, GitHub, Microsoft, Apple or email.

  2. 02

    Hand it footage.

    A link to a file, or nothing at all: it can make titles, graphics, stills and a voiceover from scratch.

  3. 03

    Say what you want.

    “A 30-second cut for LinkedIn, captions, logo at the end.” It edits, checks its work and gives you a download link.

Your agent’s setup

In Kazi (app.apifunnel.ai), open Servers and add your own server. Paste the address, Connect, then sign in.

https://cuttingfloorai.apifunnel.ai/mcp
infinite agentic memory00:00:13:00
07 / Made here

Graphics, cuts and sound, from one conversation.

A 24-minute explainer: every slide drawn with create.motion, every cut placed by an agent. This is thirteen seconds of it.

08 / Pricing

Pay for rendering.

PlanPriceIncludedOver thatFor
Trial$0 3 days60 render min—Everything, no card. About six hours of an agent editing.
Creator$49 / month300 render min$0.08 / minAbout 30 hours of editing a month. One job at a time, up to 1080p.
Pro$99 / month900 render min$0.06 / minAgents editing most days. Two jobs at once, 4K.
Enterprise$499 / seat / month5,000 min a seat$0.05 / minAgents editing all day. Priority queue, SSO, an SLA. Dedicated servers from $1,500.

A render minute is a minute of our servers working for you: exports, transcripts, graphics, voiceovers, cutouts. Planning and editing the timeline cost nothing. Yearly plans get two months free. Payments by Stripe; cancel any time.

09 / Compared

Edits, not templates.

Looking for a Remotion, Shotstack, Creatomate or JSON2Video alternative? Those render a video you have already described in code or JSON. cuttingfloor is a video editing API your agent drives call by call, looking at each result.

ToolWhat it isPick it when
cuttingfloorAn MCP server of 62 editing tools. The agent imports footage, reads the transcript, cuts, captions, mixes and checks every step with a preview.the agent decides the edit, on footage it has not seen before.
RemotionVideos written as React components and rendered from code.you want the video to be a codebase you keep and maintain.
ShotstackA cloud render API: send a JSON timeline, get a file back.you already know the edit and need it rendered at volume.
CreatomateTemplates designed in its editor, filled in through an API.one branded template needs many variations.
JSON2VideoA JSON-to-video API with built-in text-to-speech.simple videos are assembled from data.
10 / Questions

Asked.

Can Claude edit videos?

Yes, with cuttingfloor connected. Claude (web, desktop and Claude Code) calls its tools to import your footage, transcribe it, cut on words, add captions, mix the sound and export an MP4. In Claude, add it under Settings → Connectors; in Claude Code, run claude mcp add.

Can ChatGPT edit videos?

Yes, in developer mode, which lets ChatGPT use remote MCP servers. Add the cuttingfloor address under Settings → Apps → Create, sign in, and ask for the cut you want.

How do I add a video editing MCP server to Cursor?

Add cuttingfloor to ~/.cursor/mcp.json with the address https://cuttingfloorai.apifunnel.ai/mcp. Cursor opens the sign-in the first time a tool runs. The same address works in Claude, VS Code, Kazi and any client that takes remote MCP servers.

What is agentic video editing?

An AI agent edits the footage itself: it reads the material, decides the cuts, makes them with real editing tools and checks the result before it hands it over. cuttingfloor gives the agent those tools; you say what you want.

Is there a video editing API for AI agents?

cuttingfloor is one: 62 tools over MCP, from clip.add and caption.add to audio.set_master and export.video. Every edit returns a new timeline and a preview, so the agent sees what it did and can step back. When an edit needs a loop, code.run runs a script that calls the same tools.

Which agents work with it?

Any MCP client with remote servers: Kazi, Claude (web, desktop and Code), Cursor, ChatGPT in developer mode, VS Code, and anything built on the MCP SDKs.

Where does my footage go?

To our servers, for as long as you work on it. Media and renders expire after seven days, timelines after a day without use. No one else can see them.

Can I run it myself?

Yes. cuttingfloor is open source and runs in one Docker image on your own machine. The hosted version is the same image, run for you.

Do I need my own AI keys?

No. Transcription, voiceover, cutouts and graphics run on our servers. Your agent is whatever you already use.