mcp.film / Editing & Post / Kinocut
Kinocut
Guardrailed local video editing MCP — FFmpeg, receipts, quality gates, and repurposing.
What it does
Kinocut, formerly mcp-video, wraps FFmpeg and adjacent media tools behind typed, preflighted MCP operations, with a `kino` CLI and Python client over the same engine. For filmmaker agents it is the safer local post bench: inspect clips, trim and merge, burn captions, normalize audio, render variants, run release checkpoints, and fail early when an edit would produce bad media. Every operation writes a Video Receipt with input and output hashes so another agent or a human can verify what actually happened.
Connect
claude mcp add kinocut -- uvx --from kinocut kino --mcp
{
"mcpServers": {
"mcp-video": {
"command": "uvx",
"args": [
"--from",
"kinocut",
"kino",
"--mcp"
]
}
}
}
{
"mcpServers": {
"mcp-video": {
"command": "uvx",
"args": [
"--from",
"kinocut",
"kino",
"--mcp"
]
}
}
}
Sample tools
search_toolsvideo_infovideo_trimvideo_mergevideo_subtitlesvideo_normalize_audiovideo_release_checkpointvideo_repurposevideo_workflow_rendervideo_review_runhyperframes_renderField notes
Renamed: KyaniteLabs' mcp-video is now Kinocut. The slug here stays `mcp-video` so existing links keep resolving, but the project, package, CLI and docs have all moved. The old docs site kyanitelabs.github.io/mcp-video now 404s; product docs live at kinocut.dev. PyPI `mcp-video` is only a compatibility installer these days (1.6.16 pulls kinocut 1.16.1) — install `kinocut` and drive it with the `kino` CLI. The MCP server needs the `--mcp` flag (`kino --mcp`); this entry's install line omitted it before 9 Oct 2026, so bare `kino` configs copied from here will not start a server. Legacy `mcp_video` imports, MCP_VIDEO_* env vars, ~/.mcp-video data and mcp-video:// resource URIs are still honoured on the 1.14.x+ line, but that is a compatibility window, not a plan. Apache-2.0, 196 stars, Python >=3.11, v1.16.1 released Oct 5 2026, pushed Oct 9 2026, registered in the official MCP Registry as io.github.KyaniteLabs/kinocut; the README now names GitHub as canonical and the maintainer's Forgejo as a downstream mirror. Core editing needs FFmpeg on PATH and AI extras (Whisper, stems, upscalers, object matte) add gigabytes, so install only the extras you need; Hyperframes tools additionally want Node 22+. The 203-tool count is self-reported, and the README openly lists surfaces that are gated or unreleased — native desktop bundles are blocked on FFmpeg licensing, C2PA signing is development-tip only and off by default, and the trusted-execution kernel plus paid generative and dubbing plans are unshipped.
Also in Editing & Post
A local editor with no GUI and no LLM: an append-only edit graph an agent drives over MCP.
Conversational FFmpeg: trim, concat, overlay, scale, and extract frames locally.
Local MCP and CLI toolkit for agent-driven video composition, editing, analysis, and rendering.
Hosted MCP that clips, captions, dubs, reframes, and publishes long-form footage.
CLI-first video workflow runtime with an MCP surface for briefs-to-renders.
The vendor's own MCP: render JSON movies, validate them, and manage the media library.
Used in
AI Commercial Sprint · Model Shootout · Local Edit Bay · Archive To Trailer · Open Lab · Compare video and image models before spending the budget · Edit local footage without handing over the source library · Turn an archive or dailies library into a trailer · Run a hackable open film lab
Pairs well with
Quoted dubbing, subtitles, transcription and multi-voice audiobooks from one remote MCP; a human approves each price.
Official Deepgram server: Nova transcription, diarization, and audio intelligence.
Official hosted Airtable MCP: query, create, and update production-tracking bases.