v0.5.0 - Universal GPU Acceleration, MCP Progress, Cancellation & 4 New Tools

What's New in v0.5.0

🚀 MCP Protocol Super-Powers

• Live Progress Reporting (notifications/progress): Real-time percentage encoding updates streamed directly to modern MCP clients (Claude Desktop, Cursor, Antigravity).
• Instant Cancellation (AbortSignal): Child FFmpeg processes are cleanly terminated (SIGTERM/SIGKILL) immediately when a user cancels in their client.
• Timeout & Deadlock Protection: Guard against hanging processes on corrupted or network streams.

⚡ Universal Hardware Acceleration

• Automatically detects and utilizes GPU encoders across all major operating systems:
  • macOS (Apple Silicon): h264_videotoolbox, hevc_videotoolbox
  • NVIDIA GPUs (Linux/Windows): h264_nvenc, hevc_nvenc
  • Intel QuickSync: h264_qsv, hevc_qsv
  • AMD Radeon: h264_amf, hevc_amf
  • Linux VA-API: h264_vaapi, hevc_vaapi
• Runtime capability validation ensures smooth fallback to software if drivers are absent.

🛠️ 4 New Production Tools (17 Tools Total)

• burn_subtitles: Hardcodes subtitles directly onto video frames with customizable font size, colors, outline, and positioning for TikTok, Reels, and Shorts.
• extract_keyframes: Extracts representative frames across the video and generates a contact sheet grid image with inlined base64 previews for multimodal vision models.
• normalize_audio: Broadcast-standard EBU R128 loudness normalization (-16 LUFS for podcasts/shorts, -14 LUFS for YouTube) with optional speech clarity filtering.
• add_overlay: Watermark and logo overlay with customizable corner positioning, padding, scale, and opacity.

📦 MCP Resources & Prompts

• Resources: media://capabilities (FFmpeg hardware info) & media://presets (social media aspect ratios & size targets).
• Prompts: social_clip_generator & whisper_audio_prep.

🐛 Fixes & Improvements

• Fixed video A/V lip-sync issue in silence trimming by applying synchronized cuts.
• Modularized Zod schemas and expanded automated test suite to 19 passing suites.

v0.4.0 - Split Screen, Stacked Reels & Multimodal Video

Changes in v0.4.0:

• Added create_split_screen tool: stacks 2 videos vertically (1080x1920 for Reels and TikTok) or side-by-side
• Supports audio mixing (mix both, top only, or bottom only)
• Supports embedding .srt/.vtt subtitle tracks directly into the split-screen output
• Hardware acceleration enabled
• 13 total tools active and verified

v0.3.2 - Clean documentation and simplified schemas

Changes in v0.3.2:

• Simplified all tool and parameter descriptions across the server
• Removed em-dashes and promotional adjectives
• Clean, plain language for developer clarity

v0.3.1 - Enhanced Tool Documentation & Usage Guidelines

Changes in v0.3.1:

• Comprehensive behavior documentation across all 12 tools
• Explicit usage guidelines clarifying when to use each tool vs alternatives
• Complete parameter typing, defaults, and boundary descriptions for A+ TDQS ratings

v0.3.0 - Hardware Acceleration, Multimodal Vision & Aspect Reformatting

What's New in v0.3.0:

• Apple Silicon Hardware Acceleration: Automatic detection and usage of h264_videotoolbox and hevc_videotoolbox on macOS for 5x-10x faster encoding with minimal CPU load.
• Multimodal Inlined Vision: extract_thumbnail and generate_waveform return inlined base64 image data directly in the MCP response for AI vision models.
• Social Aspect Ratio Reformatting: New reformat_aspect_ratio tool transforms horizontal videos into 9:16 vertical reels (Shorts, TikTok, Reels) with aesthetic blurred background padding or smart center crop.
• Dead Air & Silence Trimmer: New trim_silence tool strips dead air and pauses from screencasts, podcasts, and recordings.
• 12 total tools active and verified.

v0.2.0 - Expanded Suite: Compression, Concatenation, Waveforms, Subtitles, & Speed

What's New in v0.2.0

We expanded ffmpeg-mcp-server from 5 tools to 10 powerful tools:

1. compress_video: Target-size and preset compressor (GitHub PR 10MB, Discord 25MB, or email) calculating optimal 2-pass bitrates.
2. concat_media: Sequential clip joining using lossless stream copy or normalized re-encoding.
3. generate_waveform: High-resolution audio visualizer generating clean PNG soundwave graphics.
4. add_subtitles: Lossless subtitle embedding for .srt and .vtt into MP4 (mov_text) and MKV containers.
5. change_speed: Playback speed adjustment (2x/4x timelapse for product demos, 0.5x slow-mo) with automatic audio pitch normalization.

All 10 Supported Tools

• media_probe
• media_convert
• trim_media
• extract_audio
• extract_thumbnail
• compress_video
• concat_media
• generate_waveform
• add_subtitles
• change_speed

v0.1.0 - Initial Release

Deterministic, rock-solid Model Context Protocol (MCP) server for FFmpeg & media processing with AI coding agents.

Features

• media_probe: Extracts complete container & stream metadata via ffprobe JSON.
• media_convert: Deterministic encoding with web_mp4, webm, and 2-pass palette gif.
• trim_media: Lossless stream copying (-c copy) or frame-accurate trimming.
• extract_audio: Audio extraction including whisper_wav (16kHz mono PCM for Whisper) & standard formats.
• extract_thumbnail: Still frame capture at custom timestamps or automatic video midpoint.
• Full TypeScript & Zod schema validation.