The "Cursor IDE" for Video Editing
AI-Native Desktop NLE for Windows with Embedded Model Context Protocol (MCP) Server
Clawvinci is the "Cursor IDE" for Video Editing.
Just as modern AI-first code editors (like Cursor or Windsurf) pair software developers with AI agents that can directly read files, inspect ASTs, suggest edits, and invoke linters inside the workspace, Clawvinci pairs video creators with AI agents that can directly read, cut, trim, arrange, color grade, and animate clips on the timeline canvas.
Built with a high-performance Rust workspace (10 modular crates) and a lightweight Tauri v2 desktop shell, Clawvinci hosts an embedded Model Context Protocol (MCP) server on http://127.0.0.1:19789/mcp. Connected AI assistants (such as Claude Desktop, Claude Code, Cursor, Codex, or local LLMs) act as collaborative co-editors—analyzing audio waveforms, cutting dead air, syncing multi-track recordings, grading colors, and generating kinetic text overlays, all preserved within a frame-accurate, shared undo/redo history.
- Frame-Accurate Multi-Track Timeline: Canvas-rendered timeline with sub-frame precision, interactive clip trimming, splitting, ripple/overwrite edits, and shared undo/redo history.
- Embedded MCP Server (53 Tools): Exposes a local HTTP/SSE Model Context Protocol endpoint (
http://127.0.0.1:19789/mcp) allowing external AI agents to programmatically inspect media, execute ripple edits, apply effects, and generate assets. - 28-Locale Localization: Full client-side internationalization across 28 languages (English, Spanish, French, German, Japanese, Simplified Chinese, Traditional Chinese, Hindi, Arabic, Russian, Korean, etc.).
- GPU Effects & Color Grading: 12 shader algorithms (tone controls, lift/gamma/gain color wheels, vignette, grain, blur, glow, curves), 3D
.cubeLUT support with tetrahedral interpolation, and styled kinetic text animations viafontdue. - High-Performance Media Engine: Built-in FFmpeg probe, decode, and encode pipeline with background waveform extraction, thumbnail caching, and bounded memory concurrency.
- Semantic Search & Transcription: Offline visual search powered by SigLIP2 (
PALMEMB1binary vector store) and on-device speech transcription (Whisper VAD & local engine, plus OpenAI BYOK Whisper REST client). - Generative AI Integration: Direct BYOK integration with generative models across modalities (OpenAI, Kling, Seedance, Fal, ElevenLabs, Suno) with automated timeline placement.
- Strictly Private & BYOK (Decision 10): Zero telemetry (
no sentry/posthog), no accounts or subscriptions (no clerk), and no cloud middleman (no api.palmier.io). All AI calls go directly to the provider with your own keys or run locally on your device. - Universal Project Compatibility: Native support for the byte-compatible
.palmierproject package format, alongside export interchange to DaVinci Resolve (FCPXML 1.10–1.14) and Premiere Pro (XMEML 4).
When Clawvinci is running, it hosts an embedded MCP server at http://127.0.0.1:19789/mcp. You can connect your favorite AI assistant to edit your timeline directly:
claude mcp add --transport http clawvinci http://127.0.0.1:19789/mcpAdd the following to your Cursor MCP settings (%USERPROFILE%\.cursor\mcp.json):
{
"mcpServers": {
"clawvinci": {
"type": "http",
"url": "http://127.0.0.1:19789/mcp"
}
}
}Add the following to your Claude Desktop configuration (%APPDATA%\Claude\claude_desktop_config.json):
{
"mcpServers": {
"clawvinci": {
"type": "http",
"url": "http://127.0.0.1:19789/mcp"
}
}
}codex mcp add clawvinci --url http://127.0.0.1:19789/mcpClawvinci is organized as a modular Rust workspace inside windows/src-tauri/crates/:
| Crate | Purpose | Status |
|---|---|---|
clawvinci-model |
21 domain models (Timeline, Track, Clip, Timecode, ProjectFile), serialization, and byte-compatible .palmier package format. |
✅ 100% Complete |
clawvinci-media |
FFmpeg-based video/audio probe, frame decoding, thumbnail cache, waveform generation. | ✅ 100% Complete |
clawvinci-timeline |
NLE editing primitives, ripple/overwrite engine, track/clip operations, command undo/redo history. | ✅ 100% Complete |
clawvinci-render |
FramePlan builder, compositor, 12 GPU effects, 3D LUT parser, text rasterization & animators. |
✅ 100% Complete |
clawvinci-export |
Render-to-file batch export, FCPXML 1.10–1.14, Premiere XMEML 4, and export queue. | ✅ 100% Complete |
clawvinci-mcp |
Embedded HTTP/SSE MCP server exposing 53 timeline editing tools to external AI agents. | ✅ 100% Complete |
clawvinci-audio |
Real-time peak/RMS metering, cross-correlation audio sync, silence/dead-air detection, tempo/beat detector, VAD. | ✅ 100% Complete |
clawvinci-search |
Semantic visual search (SigLIP2 ONNX, PALMEMB1 store) and transcript search. |
✅ 100% Complete |
clawvinci-gen |
Direct BYOK generative AI clients (video, image, voice, music) with timeline insertion. | ✅ 100% Complete |
clawvinci (Tauri App) |
Application shell, IPC bridge, window management, persistent settings, Home hub, and Web UI. | ✅ 100% Complete |
- Windows 10 (22H2+) or Windows 11 (x86_64).
- Visual Studio C++ Build Tools with Windows 10/11 SDK.
- Rust (
rustup default stable-x86_64-pc-windows-msvc). - Node.js (v20+) and npm.
- FFmpeg on system
PATH.
-
Clone the repository:
git clone https://github.com/HasNetwork/Clawvinci.git cd Clawvinci
-
Install frontend dependencies:
cd windows npm install
-
Run tests across all Rust crates:
cd src-tauri cargo test --workspace
-
Verify lints with zero warnings:
cargo clippy --workspace -- -D warnings
-
Launch Clawvinci in desktop development mode:
cd .. npm run tauri dev
-
Build release package (NSIS installer):
npm run tauri build
The .agents/ folder serves as the single source of truth for the project:
- Developer Handover Manual:
.agents/handover.md - Master Architecture Plan:
.agents/PLAN.md - Implementation History:
.agents/HISTORY.md - Phase 13.0 Polish Log:
.agents/HISTORY/13-0-polish.md - Coding Rules & Invariants:
.agents/RULES.md
Clawvinci is licensed under the GNU General Public License v3.0 (GPL-3.0-only). See LICENSE for details.