Skip to content

Releases: AISEO-VirtualabDigital/mcp-video-editor

v1.2.0 — write path, typing, onboarding

Choose a tag to compare

@AISEO-VirtualabDigital AISEO-VirtualabDigital released this 29 Oct 16:02

No new tools (still 59). Saves are cheaper, types are honest, and there is a real onboarding guide.

Performance

  • CapCut material_type resolves through one Map instead of a find per clip — O(clips + materials).
  • Projects serialize once per save; an EBUSY retry re-runs only the file write.
  • Atomic write narrowed: temp file opened wx, single handle, close, then rename — only the rename touches draft_content.json, and a failed write removes its temp file.
  • Dry-run preview snapshots are opt-in (MCP_VIDEO_DRY_RUN_SNAPSHOT=1). Every dry-run used to serialize and write a second full copy of the project.

Correctness

  • OpenCut writes now preserve unmodelled vendor fields at root/track/clip/material/text/effect level. CapCut already did.

Typing

  • New src/adapters/raw.ts (RawBag, isRecord, toRawBag, typed readers). Type guards replace casts in validator.ts and server.ts. Explicit casts 12 → 6.

Docs & tooling

  • New docs/GETTING-STARTED.md: install on Windows/Unix, manual registration for all 8 clients, runtime contract, troubleshooting.
  • New npm run smoke — 37 end-to-end checks in a throwaway sandbox, including the full path-guard refusal matrix; npm run verify runs typecheck + smoke.

Verified: tsc --noEmit, build, 37/37 smoke checks, 12/12 xmeml assertions, 8/8 vendor round-trip assertions.

v1.1.0 — 59 tools, effects catalog, voice, beta Premiere/Filmora export

Choose a tag to compare

@AISEO-VirtualabDigital AISEO-VirtualabDigital released this 29 Oct 15:43

Added

  • Effects catalog — list_effects, get_effect over 20 curated effects (face-tracked, kinetic-text, color-grade, motion, sticker, ambient, checklist) with faceSafe + timingHint.
  • apply_face_effect — applies a catalog effect over a timeline window, validates the id (did-you-mean), rejects end <= start, warns when the effect is not face-safe.
  • generate_voice — MP3 voiceover via OpenAI or ElevenLabs.
  • motion render engine — third engine beside Remotion and HyperFrames.
  • Shot library — list_shots, get_shot.
  • export_timeline_xml (BETA) — FCP7 XML for Premiere/Filmora. Untested in-app.
  • Qwen Code + ZCode CLI client configs.

Fixed

  • install-unix.sh wrote Windows paths into macOS/Linux client configs.
  • applyFaceEffectSchema was dead code; it now backs a real tool.
  • xmeml builder referenced <file> ids without defining them, and left & unencoded in pathurl.

Not supported

Premiere and Filmora remain unsupported editors. The XML export is beta and has never been imported into a real install — the XML is unit-tested and parses, nothing more.