AGENTCONN

// Skill profile

FFmpeg Skill

Framework Agnostic Beginner Creative Open Source

kajisho5/ffmpeg-skill is an open-source skill and MCP server that gives AI coding agents access to 42 FFmpeg-based video and audio processing tools. The skill exposes operations like trimming, concatenation, format conversion, thumbnail extraction, audio normalization, subtitle burning, resolution scaling, and codec transcoding through a structured tool interface that any MCP-compatible agent can call. All processing runs locally — no cloud uploads, no API costs, no file-size limits beyond disk space. MIT licensed, 5,478 GitHub stars with +168/day velocity as of November 2026. The MCP server wraps FFmpeg's complexity behind typed tool definitions with validated inputs and human-readable error messages.

Input / Output

Accepts

video-file audio-file processing-params

Produces

processed-video processed-audio thumbnail metadata

Overview

FFmpeg is the Swiss Army knife of multimedia processing, but its command-line interface is notoriously hostile — hundreds of flags, codec-specific options, and filter graphs that look like line noise. ffmpeg-skill wraps all of that behind 42 typed tool definitions that an AI agent can call directly. Instead of an agent trying to construct an FFmpeg command string (and getting it wrong half the time), it calls a tool like trim_video(input, start_time, end_time) and gets a processed file back.

The skill ships as both a SKILL.md instruction bundle (for Claude Code and compatible harnesses) and an MCP server (for any MCP-compatible client). All processing runs locally using the system’s FFmpeg installation — no cloud uploads, no API costs, no file-size restrictions beyond available disk space. This makes it particularly useful for workflows involving large video files, proprietary content, or air-gapped environments.

Key Capabilities

42 processing tools: Covers the most common multimedia operations — video trimming, concatenation, speed adjustment, resolution scaling, format conversion, codec transcoding, thumbnail extraction, GIF creation, audio extraction, audio normalization, volume adjustment, subtitle burning, watermark overlay, and more. Each tool has typed parameters with validation.

MCP server: Runs as a standard MCP server that any compatible client can connect to. Tool definitions include parameter schemas, descriptions, and examples. Error messages are human-readable, not raw FFmpeg stderr.

Local-only processing: No files leave the machine. Processing speed depends on local hardware — GPU acceleration is supported where FFmpeg and the system support it. No API keys, no usage limits, no per-minute charges.

Composable operations: Tools can be chained — trim a video, then scale it, then add a watermark, then convert to GIF. The agent orchestrates the pipeline; each tool handles one step cleanly.

Metadata extraction: Tools for reading video/audio metadata — duration, resolution, codec, bitrate, frame rate — so agents can make informed decisions about processing parameters.

Use Cases

  • Content pipeline automation — agents that process video uploads (transcode, generate thumbnails, create preview clips) as part of a content management workflow
  • Video editing assistance — an agent that takes natural-language editing instructions (“trim the first 30 seconds, add subtitles, export as 720p MP4”) and executes them
  • Batch processing — convert, resize, or normalize a directory of media files with a single agent instruction
  • Podcast production — extract audio, normalize levels, trim silence, and export in multiple formats
  • Social media repurposing — take a long video and produce platform-specific cuts (vertical for Reels/TikTok, square for Instagram, horizontal for YouTube)

Considerations

Requires FFmpeg installed on the system (the skill does not bundle it). Processing speed is bound by local hardware — large files or complex filter graphs may take time. The tool set covers common operations but does not expose FFmpeg’s full filter graph syntax; highly custom processing may still require manual FFmpeg commands. GPU acceleration depends on the system’s FFmpeg build supporting hardware codecs.

Who It’s For

Developers building media-processing pipelines with AI agents. Content teams that want agents to handle video/audio processing tasks. Anyone who uses FFmpeg regularly and wants their coding agent to handle it instead. Teams processing media locally for privacy, compliance, or cost reasons.

Tags

#skills #ffmpeg #video #audio #multimedia #mcp #local-processing #creative #coding-agents

Compatible Agents

AI agents that work well with FFmpeg Skill.

Similar Skills