Kinocut 1.8.0

Kinocut 1.8.0.

Thanks to @betsmayank for the Hyperframes MCP no-TTY initialization fix (#361).

Native MCPB remains staged/local-only and is not published as an external artifact.

Kinocut v1.7.0

Kinocut is the new canonical identity for MCP Video: the same trusted local video execution layer, now with stable kinocut/kino entry points and a compatibility-preserving upgrade path.

Added

• Kinocut is now the canonical Python distribution and import package, with kino and kinocut CLI commands alongside the preserved mcp-video command.
• The kinocut npm package provides a thin uvx launcher for the kino CLI, and the MCP Registry entry now publishes under the immutable io.github.KyaniteLabs/kinocut identity.
• The canonical skills/kinocut agent skill teaches both MCP and CLI workflows; skills/mcp-video remains as a compatibility pointer.

Changed

• The project, repository links, documentation, package metadata, and public discovery surfaces now use the Kinocut name and kinocut.dev home.
• The implementation package moved from mcp_video to kinocut. New Python integrations should import from kinocut.
• Current documentation, examples, workflows, active plans, and research guidance now use canonical Kinocut names, paths, install commands, and the verified 135-tool/114-command surface. Dated evidence retains its historical identity with explicit snapshot notices.
• The documentation map separates current operating guidance from historical audits, proofs, and handoffs; local Markdown links and canonical naming are enforced by public-surface tests.

Compatibility

• mcp-video==1.6.1 is a metadata-only upgrade shim that installs kinocut==1.7.0 and forwards every optional extra. Existing mcp_video imports, mcp-video commands, MCP_VIDEO_* environment variables, ~/.mcp-video data, mcp-video:// resource URIs, and receipt keys remain supported through at least Kinocut 1.8.x.
• Clean installs and in-place upgrades from mcp-video==1.6.0 are exercised as release gates, including Python imports, all three CLI entry points, package metadata, and uninstall behavior.

Fixed

• MCP Registry metadata now uses the registry-supported GitHub mirror URL, while Forgejo remains canonical everywhere else. The publish workflow also supports a registry-only recovery dispatch, so a downstream registry rejection can be repaired without attempting to re-upload an immutable PyPI release.
• The MCP stdio handshake now reports kinocut as the server name instead of the retired project name.
• Contribution, testing, licensing, design-standard, integration, workflow, and agent setup docs now match the current package layout and release behavior; the Remotion removal version is corrected to v1.3.1.

Install or upgrade

pip install --upgrade kinocut
# Existing users can also upgrade without changing integrations:
pip install --upgrade mcp-video

v1.6.0

MCP Video 1.6.0 expands the project from individual editing tools into a guarded execution layer for agent-driven video workflows.

Highlights

• Added a local workflow engine with validate, plan, render, inspect, resume, variants, cleanup, provenance receipts, and fail-closed workspace confinement across MCP, CLI, and Python.
• Added the dedicated rescue pipeline for source-backed diagnosis, reviewable repair plans, approved rendering, verified master and sharing outputs, resumable intermediates, and hashed receipts.
• Added post-rescue planning and verification surfaces for semantic timelines, timeline edits, subject-aware transforms, restoration, composition, creative autopilot, and explicit remote-egress contracts.
• Expanded compositing with workflow-safe layer stacks, transforms, masks, timing windows, full-canvas blend modes, rotation and pivot controls, richer receipts, and source/output hashes.
• Added audio ducking, LUT-based color grading, progress notifications, bilingual common errors, and Spanish README guidance.

Production Fixes

• Audio-only WAV normalization and muxing now use audio-aware probes without false video-stream failures.
• Hyperframes relative project paths are normalized once instead of duplicating the project directory.
• Audio composition accepts supported WAVE_FORMAT_EXTENSIBLE PCM WAV files through canonical internal decoding.
• Technical and design quality checks share explicit saturation and contrast metrics; --fail-on-warning now exits nonzero whenever the result fails.

Safety And Compatibility

• Workflow paths, writes, sources, approvals, dependencies, resumptions, resource caps, and engine failures fail closed with structured errors and privacy-safe receipts.
• Rescue remains local-only and source-immutable. It does not download models or silently invoke cloud services.
• Existing MCP, CLI, and Python APIs remain compatible. Intentional 24 fps delivery behavior is unchanged.
• Portability is exercised against pinned FFmpeg 6, 7, and 8 builds.

Upgrade with:

pip install --upgrade mcp-video

Full details are in the changelog.

v1.5.1

Patch release for Glama/package-builder compatibility.

Fixed:

• Removed the unpublished meltysynth dependency from optional audio extras so uv, Glama, and package metadata validators can resolve mcp-video.
• Updated doctor/runtime guidance for MeltySynth to avoid recommending a nonexistent PyPI package.
• Updated repository readiness audit after the intentionally removed /explainer-video Dependabot path.
• Added regression coverage so optional extras do not advertise unpublished dependencies.

Proof before release:

• Local Glama-like resolver path now builds and imports mcp-video 1.5.1.
• Full local tests passed: 1275 passed, 12 skipped, 16 warnings.
• PR #346 checks passed.
• Master CI and integration smoke passed.
• Built wheel metadata contains no meltysynth requirement.

v1.5.0

Highlights

• Added Video Receipts as the review trail for agent-created or agent-edited media.
• Added receipt-backed confidence baseline and local repurpose-package workflows.
• Added confidence benchmark and adversarial certification scripts for local readiness proof.
• Hardened the explainer workflow into a 10-stage lane with audio normalization, transition audit records, quality report, and a real min_score=50 release checkpoint.

Fixes

• Removed the stale /explainer-video npm Dependabot entry so dependency automation can run against real package paths.
• Made corrupted media probing fail clearly as InputFileError.
• Fixed the confidence benchmark so stale receipts cannot hide a failed workflow run.

Verification

• PR #342 merged with Hosted PR checks, agent-law, and CodeRabbit passing.
• PR #343 proved Dependabot automation is alive after the config fix.
• PR #344 aligned pyproject.toml, mcp_video.__version__, and server.json at 1.5.0.
• Local package build and twine check passed for mcp_video-1.5.0.

v1.4.1

Highlights

• Added preflight guardrails across high-risk video/audio edit paths: filters, merge, add-audio, overlay, watermark, chroma key, text animation, grid layout, and split screen.
• Refreshed docs and metadata for the current 119-tool guardrailed surface.

Fixes

• Fixed base installs without optional NumPy extras so public audio facades and CLI/doctor imports stay available.
• Added pure-Python fallback coverage for the newer DSP symbols and typed validation errors for invalid fallback synthesis frequencies.
• Hardened no-NumPy regression coverage so the fallback path is exercised even if NumPy was already imported.
• Shortened MCP registry metadata description so the release workflow passes the live registry schema.

v1.4.0

What's new in 1.4.0

New features

• Hyperframes 0.5 orchestration — snapshots, inspection, metadata, catalog, website capture, local TTS, transcription, background removal, diagnostics, benchmarks, and richer render controls
• Video repurposing helpers — dry-run manifests and platform-ready variants with thumbnails, storyboards, and optional release checkpoints
• Real color extraction — analyze_video(include_colors=True) and video_info_detailed now return actual dominant colors instead of placeholder values

Changed

• Pipeline cleanup OSErrors are now surfaced as warnings instead of being silently swallowed

Fixed (30+ hardening fixes from PRs #263–#292)

• Fixed green-cast video effects and 24-bit WAV handling
• Design quality probe now returns None on analysis failure instead of sentinel values that produced perfect scores
• Comprehensive input validation across audio, video, Hyperframes, timeline, template, batch, subtitle, and quality comparison paths — invalid inputs now fail loudly before reaching FFmpeg/rendering
• Hyperframes render artifacts reported correctly by format, pipelines fail on missing renders

Full test suite: 1173 passed, 10 skipped

Full Changelog: v1.3.10...v1.4.0

v1.3.10

Fixed

• Improved mcp-video doctor guidance on Python 3.13+ for missing Real-ESRGAN/BasicSR so users see the BasicSR build guard, OpenCV fallback path, and Python 3.11/3.12 backend option.

Verification

• PR #259 passed hosted PR checks before merge.
• master CI and integration smoke passed after merge.
• Local Python 3.13 doctor JSON smoke verified the new hint before release.

v1.3.9

Fixed

• Fixed mcp-video[all-ai] and mcp-video[upscale] installs on Python 3.13 by guarding Real-ESRGAN/BasicSR behind the Python versions where BasicSR still builds.
• Kept the current-Python all-AI/upscale path installable through the OpenCV upscaling fallback.

Verification

• PR #258 passed hosted PR checks before merge.
• master CI and integration smoke passed after merge.
• Clean Python 3.13 local wheel install with mcp-video[all-ai]==1.3.9 succeeded, and the OpenCV upscale fallback produced a 1280x720 output from the smoke fixture.

v1.3.8

Fixed

• Fixed AI scene-detection JSON output so perceptual-hash differences serialize as standard JSON numbers.
• Added torchcodec to stem-separation extras and diagnostics so Demucs output works with current TorchAudio save behavior.

Verification

• PR #257 passed hosted PR checks before merge.
• master CI and integration smoke passed after merge.
• Local release build, twine check, wheel metadata inspection, source CLI AI-scene JSON smoke, repo readiness audit, and the full local suite were verified before release.