FFmpeg as a service, without handing anyone a shell.

KinoPipe is a video editing tool for AI agents. Underneath, it runs FFmpeg 9 on managed workers and exposes it as typed video editing operations. Your product or AI agent sends a validated request and receives a downloadable file: no FFmpeg install, no command strings, no video infrastructure.

45 typed tools

Trim, resize, crop, concat, split, watermark, subtitles, audio and more, each validated by schema.

FFmpeg 9 workers

Compiled from source with libx264, libvpx, libass and tuned encoder defaults.

Async jobs

Queued processing with progress, typed failure codes, idempotency keys and signed webhooks.

Hosted MCP server

OAuth 2.1 remote MCP so Claude, Codex or ChatGPT can edit video directly.

A shell wrapper trusts the caller. KinoPipe validates the request.

Most hosted FFmpeg products accept command strings. That works until the caller is a language model: one hallucinated flag and you are debugging injection, runaway encodes or malformed outputs. KinoPipe validates operation types, dimensions, time ranges and input references before a job is queued.

RAW FFMPEG: WHAT A SHELL WRAPPER ACCEPTS
ffmpeg -i input.mp4 \
  -vf "scale=1080:1920:force_original_aspect_ratio=increase,\
       crop=1080:1920" \
  -c:v libx264 -preset fast -crf 23 \
  output.mp4
# ...or anything else the model writes.
KINOPIPE: WHAT YOUR AGENT SENDS
POST /api/v1/tools/crop-video
{
  "inputs": [{ "id": "main", "url": "https://…/video.mp4" }],
  "options": { "aspect_ratio": "9:16", "focus": "center" }
}
// Validated, bounded, queued, or rejected with a typed error.

8.2s

p50 render for a 1080p H.264 edit

100%

success rate across 325 benchmark runs

0

shell commands accepted

FFmpeg 9 render benchmark, August 2026: cold, warm and burst phases on production recipes. Read the full report: CPU vs GPU, cold starts, cost per job.

Call it from code, workflows or agents.

Prebuilt tool endpoints
45 focused endpoints for common edits: POST /api/v1/tools/compress-video and friends. The fastest path.
Composable jobs
Combine up to 12 operations on up to 6 inputs in one recipe with POST /api/v1/jobs, with signed uploads for local files.
Remote MCP
Connect once with OAuth 2.1 and your agent discovers every editing tool with typed schemas, no API key to paste.

Honest comparisons before you choose.

Running FFmpeg yourself means compiling and patching it, sizing workers, securing input URLs against SSRF and babysitting a queue (the commands themselves are the easy part: see the FFmpeg command reference). If you would rather manage that surface, or need a different editing model, these comparisons say when to pick someone else.

FFmpeg as a service, answered

Hosted FFmpeg workers behind an HTTP API, so you never install, compile or scale FFmpeg yourself. KinoPipe runs FFmpeg 9 on managed workers and adds a typed request layer designed for AI agents and automations.

No, by design. KinoPipe validates every request against bounded operation schemas and never accepts shell commands. If you need the full FFmpeg command surface, compare KinoPipe with Rendi.

Trim, resize, crop, compress, convert, concat, split, watermark, picture-in-picture, subtitles, audio extraction and mixing, thumbnails and GIF. Outputs: MP4, WebM, MP3, WAV, JPEG, WebP and GIF.

One credit covers one second of worker time or 6 MiB of output; the higher rounded-up total applies, and reading tools are billed per minute of media instead. The free plan includes 100 credits with no credit card, and failed jobs are refunded automatically.

Through the hosted MCP server at https://kinopipe.com/mcp with OAuth 2.1 (one command in Claude Code or Codex) or through the REST API with an API key for server-side use.
Run your first FFmpeg job in the browser.
100 free credits, no credit card. Upload a file, pick an operation, download the result.
Start for free