Here’s something that’s been saving me a ton of time: connecting Antigravity directly to DaVinci Resolve Studio so the agent can organize media, build timelines, and strip dead air before I ever touch the UI.
The piece that makes it possible is davinci-resolve-mcp, an open-source server built on Anthropic’s Model Context Protocol (MCP) that exposes Resolve’s scripting API as executable tools for autonomous AI agents. Once wired up, you describe the edit goals in natural language and the agent handles repetitive assembly.
Architecture: Antigravity AI orchestrates DaVinci Resolve Studio through davinci-resolve-mcp via local JSON-RPC and scripting APIs.
1. One-Time Setup (Under 3 Minutes)
Enable External Scripting in Resolve
Open DaVinci Resolve Studio, go to Preferences > System > General, and set External scripting using to Local.
If you’re on Resolve 21+, you can also go to File > Setup AI Assistants > Antigravity to enable it from there.
Install the MCP Server
With Resolve open, run the interactive installer:
npx davinci-resolve-mcp setup
It auto-detects Antigravity and writes the server config to:
~/.gemini/antigravity/mcp_config.json
or if you use the global config path:
~/.gemini/config/mcp_config.json
Verify the Connection
Restart Antigravity, then run a quick health check:
npx davinci-resolve-mcp doctor
Or ask the agent directly:
Run resolve_control(action="get_version") and tell me if it connected successfully.
If it comes back with your Resolve version number, you’re good.
2. Three Workflows Worth Setting Up
Bin and Timeline Scaffolding
Creating the same folder structure and track layout for every project is mindless work. Tell the agent to do it:
Scan my footage in ./raw_footage, build a Media Pool bin structure with folders for RAW_VIDEO, AUDIO_ASSETS, and TIMELINES, then create a 24fps timeline called "Master_Edit" with 3 video tracks (V1_A_Roll, V2_B_Roll, V3_Titles) and 3 audio tracks (A1_Dialogue, A2_Music, A3_SFX). Add color-coded chapter markers at the timecodes I'll give you.
The agent uses media_pool and timeline MCP tools to build this out without you clicking around.
Batch Clip Analysis Before You Cut
Before pulling anything into the timeline, have the agent analyze your raw clips first. Open the control panel dashboard:
Run resolve_control(action="open_control_panel") and open the local dashboard.
Then kick off transcription and visual analysis on your raw footage:
Run media_analysis on 01_RAW_VIDEO with transcription and vision enabled. Tag selects with I and rejects with X based on the transcript and visual quality.
This runs local Whisper on the dialogue and samples frames without touching your camera originals. You review selects in the browser at http://127.0.0.1:8765 before anything goes into the timeline.
Dead-Air Removal and B-Roll Assembly
Once you’ve reviewed selects, you can have the agent plan the rough cut:
Use edit_engine(action="plan_tighten") to find and cut silent pauses below -45dB in the dialogue clips. Then use plan_selects to layer B-roll clips onto V2 (video-only, no audio) so the scratch audio on the camera clips stays untouched.
Every timeline change saves a snapshot in the Control Panel’s Timeline History, so you can roll back any step that didn’t land.
3. Starter Prompt to Copy
Paste this into Antigravity at the start of any new project:
Connect to my active DaVinci Resolve Studio project via davinci-resolve-mcp and open the local Control Panel.
1. Use ffmpeg to strip leading and trailing silence below -45dB from the music tracks in ./music and export clean WAVs to ./music/trimmed_ready.
2. In the Media Pool, create bins for 01_RAW_VIDEO, 02_AUDIO_ASSETS, and 03_TIMELINES, and import the media.
3. Create a 24fps timeline named "Master_Edit" with 3 Video tracks (V1_A_Roll, V2_B_Roll, V3_Titles) and 3 Audio tracks (A1_Dialogue, A2_Music, A3_SFX).
4. Run media_analysis on 01_RAW_VIDEO with transcription and vision enabled so I can review selects in the Control Panel before running edit_engine(action="plan_tighten").
The Split That Makes This Work
Think of it this way: Antigravity handles everything that doesn’t require your eyes. Bin organization, metadata tagging, silence detection, rough assembly, render queue. You handle everything that does: color work, mixing levels, motion graphics timing, final pacing.
Multicam syncing, Fairlight sidechain compression, node-based color grades — that stuff stays in the UI where it belongs. The MCP layer just means you’re not spending the first 45 minutes of every session doing setup work you’ve done a hundred times.
Frequently Asked Questions
Does this require DaVinci Resolve Studio or does it work on the free version?
DaVinci Resolve Studio is required. Blackmagic Design restricts external scripting API access (ResolveScript) exclusively to the paid Studio license.
Is footage or audio sent to the cloud?
No. All media analysis runs 100% on your local machine using local Whisper for speech-to-text transcription and local FFmpeg for audio processing. Raw media files never leave your system.
How do I undo timeline changes executed by the agent?
Every batch edit executed through edit_engine logs a snapshot in the local Control Panel Timeline History (http://127.0.0.1:8765), allowing one-click rollbacks. Standard Cmd+Z / Ctrl+Z within DaVinci Resolve also works for immediate step-by-step undos.
Where can I explore related developer automation?
Check out my Antigravity developer workflow setup and explore more open-source tooling under Projects.
Leave a comment