Skip to content
ML2

ffmpeg-skill: a local video editor for your coding agent

Added Copyable prompt below

Published Last verified Released

Our verdict

The most complete way we've seen to hand routine video edits to a coding agent without uploading footage anywhere: trims, silence removal, burned-in captions, 16:9 to 9:16 reframes, loudness and per-platform exports, each followed by a check the agent can read. It fixes the usual failure where an agent guesses frame rates and calls a broken file done. It only executes edits; deciding what to cut or what the captions say is still your job. Releases ship almost daily (v2.17.3 on 2026-10-28), so pin a version for anything you automate.

Who it's for
Creators, indie makers and developers who cut talking-head clips, demos or podcasts for Reels, TikTok, Shorts, YouTube or LinkedIn and already use a coding agent
Leverage
Repeatable cut, caption and export jobs done by your agent on your own machine, with a verification step instead of eyeballing every file
Effort
Low
Date added

What it does

ffmpeg-skill (kajisho5/ffmpeg-skill, MIT, npm package ffmpeg-skill) is an Agent Skill plus 42 Python tools that call your local ffmpeg and ffprobe: cut, join, silence removal, fit to a duration or aspect ratio, captions with karaoke highlighting, overlays and motion graphics, HDR to SDR and LUTs, audio clean-up, sync with drift correction, multicam, loudness, delivery checks and batch folders. The skill teaches the agent a fixed order (probe, edit losslessly where possible, check, verify), and look.py produces a contact sheet the agent inspects before it reports back. Platform templates cover TikTok, Reels, Shorts, YouTube, X, LinkedIn, Facebook and podcasts. Every tool is also an MCP tool; the server lists a core 12 by default to keep context small (FFMPEG_SKILL_MCP_FULL=1 lists all 42). It needs FFmpeg 5.0+ (6.1+ recommended), Python 3.10+ with the standard library only, and Node 16+ for the installer. No account and no API key.

How builders use it

  • Run npx ffmpeg-skill to install the skill for Claude Code (add --cursor, --codex or --all for other agents), then npx ffmpeg-skill doctor to see which tools your FFmpeg build supports.
  • Ask in plain language, for example keep two time ranges of interview.mp4 and make a 60-second 9:16 clip for Reels; the agent probes, cuts, reframes, exports and checks the file.
  • Use the platform templates to render one edit for several destinations at once.
  • For other MCP clients, point the client at the bundled mcp/server.py with python3.

What to expect

Platform-ready clips with captions and correct loudness, made on your own machine by the agent you already use, with a report of what the final file actually contains.

Copy this: Install + first edit

# Official source: github.com/kajisho5/ffmpeg-skill (MIT). Needs ffmpeg 6.1+, Python 3.10+, Node 16+.
npx ffmpeg-skill            # Claude Code; add --cursor, --codex or --all
npx ffmpeg-skill doctor     # check your FFmpeg build per tool

# Then ask your agent:
Take interview.mp4, keep 0:45-3:10 and 5:00-6:30, make it exactly 60 seconds at 9:16 for Reels with burned-in captions from cues.txt, normalise loudness, then check the final file and show me the contact sheet.

# Other MCP clients:
{"mcpServers": {"ffmpeg-skill": {"command": "python3", "args": ["~/.claude/skills/ffmpeg-skill/mcp/server.py"]}}}

Related discoveries

See the newest picks →

Get the best new A.I. picks in your inbox

The best new agent workflows and tools we’ve vetted, with a clear verdict on each. Short, occasional, useful.

More in Creative & Media →

Free. No spam. Unsubscribe anytime.