Pack

DaVinci Resolve MCP skills

samuelgursky/davinci-resolve-mcp

Agent skills shipped with a DaVinci Resolve MCP server, covering grading, editing, conform, Fairlight audio, delivery QC, Fusion, media-pool ingest, footage analysis, rough cuts and dead-air removal.

View on GitHub
01Author
samuelgursky
03Stars
3,291 · as of Oct 3, 2026
04Last updated
Sep 30, 2026
05License
MIT
06Works with
Claude Code · Codex · Cursor
07Models
DaVinci Resolve · FFmpeg · Whisper
08Content language
English
09Domains
Filmmaking
10Stages
Editing · Audio
SC 01Overview

What it does

The domain skills mostly act as routers: for one post-production job they name the Resolve MCP tools to call, the repo guides to read and the traps to avoid, either in a running Resolve or offline on project files. They cover color grading, cutting and variants, conform and relink, Fairlight audio and loudness, render setup and deliverable QC, Fusion comps, media-pool ingest and read-only footage analysis, plus an index skill and a session-start check. Two workflow skills are fuller recipes: resolve-rough-cut builds a gapless vertical rough-cut timeline from a folder of raw clips, and resolve-tighten-recording removes dead air from one long take into a variant timeline after a review step. It needs DaVinci Resolve with scripting enabled (Studio directly, the free edition only through an in-app bridge), Python and ffmpeg, and the MCP server installs with one npx command.

Included skills

  • resolve-mcp
  • resolve-color
  • resolve-edit
  • resolve-conform
  • resolve-audio
  • resolve-delivery
  • resolve-fusion
  • resolve-media-pool
  • resolve-media-analysis
  • resolve-rough-cut
  • resolve-tighten-recording
  • resolve-session
SC 02Install

How to install it

  • 01From the README

    The install command quoted from the repository README.

    npx davinci-resolve-mcp setup
  • 02skills CLI, for Claude Code, Codex and others

    The open skills CLI from Vercel Labs installs a skill from a GitHub repo into the folder each agent expects. It asks which skills and which agents you want, and it installs for the current project unless you add the -g flag, which installs for your user account. Add --list to see what a repo contains before you install anything.

    npx skills add https://github.com/samuelgursky/davinci-resolve-mcp/tree/main/.agents/skills
  • 03Claude Code, skills folder

    Copy the skill's folder, the one that holds SKILL.md, into ~/.claude/skills/ to use it in every project on your machine, or into .claude/skills/ in a repository to use it there and share it by committing it. Claude Code picks up changes to these folders during a session. If the folder did not exist when the session started, run /reload-skills.

  • 04Claude Code, plugin marketplace

    Some repos publish a plugin marketplace, a .claude-plugin/marketplace.json file. Add it in Claude Code, then install a plugin from it with /plugin install plugin-name@marketplace-name. Plugins can bundle hooks and MCP servers along with skills, so read the details panel before you confirm.

    /plugin marketplace add samuelgursky/davinci-resolve-mcp
  • 05Codex, skills folder or $skill-installer

    Codex reads skills from .agents/skills in your working folder and its parent folders up to the repository root, and from $HOME/.agents/skills for every repository. Copy the skill's folder into one of them, or run $skill-installer inside Codex and ask it to install from the repo. Codex detects new skills automatically and needs a restart only if one does not appear.

The first command is quoted from the author's README. The others are the standard ways to install a skill from a GitHub repository. They are templates and may not apply here, for example if the repo has no plugin marketplace, so follow the README first.

SC 03Guide

How to get the most from it

Good for

  • Editors who want an agent to handle grading, conform, Fairlight loudness checks and deliverable QC inside a running Resolve project, or offline on project files.
  • Short-form creators who need a gapless vertical rough-cut timeline assembled from a folder of raw phone or behind-the-scenes footage.
  • Podcast, tutorial and screen-recording makers who want pauses removed from one long take as a separate variant timeline.
  • Users who want guarded changes to Resolve, such as dry-run plans, confirmation tokens and read-only media analysis, instead of free-form scripting.

Watch out for

  • Needs DaVinci Resolve with scripting enabled: Studio connects directly, while the free edition relies on an in-app bridge that the README treats as a 21.0.x path because 21.1 moved Python scripting to Studio. Python 3.10+ and ffmpeg are also required.
  • Several skills are thin routers into guides and kernel docs inside the repo and mention craft skills (colorist, editor, online-editor) that the repo does not ship, so use them from a checkout or through the server's knowledge tool.
  • Actions change real projects and the scripting API differs between Resolve patch releases, so review each plan and check the result it reports back; the tighten workflow archives the original timeline instead of overwriting it.

Typical workflow

01Step
Run npx davinci-resolve-mcp setup, then in Resolve Studio set Preferences > General > External scripting using to Local and open your project.
02Step
Describe the job in plain words, such as a rough cut of a footage folder for Reels or removing dead air from a tutorial; the matching skill is picked from its description.
03Step
For a rough cut, the agent probes the clips with ffprobe, imports them into a bin for the episode, runs read-only analysis and sets the timeline frame rate, then asks you to set the playback frame rate in the project settings because the API cannot.
04Step
For a tighten, it transcribes the take (local Whisper), produces a dry-run plan and shows the number of cuts, the time removed and the longest cuts; after you confirm, it builds a variant timeline.
05Step
The agent checks the result for gaps and overlaps, expected duration and audio present, and reports what it found.
06Step
Continue with the grading, audio or delivery skills, then render.
SC 04Fit

Where it fits in a film pipeline

The production stages this skill covers, and where Pencil0 picks up.

SC 05Related

Related skills

  • Pack

    OpenMontage

    calesthio/OpenMontage

    An open-source repo in which a coding agent directs video production through staged pipelines, with a large set of skills on video-model APIs, music, speech, editing and composition.

    • Pipeline
    • Generation
    • Editing
    62.4kGitHub starsUpdated Sep 2026
  • video-use

    browser-use/video-use

    A skill that lets a coding agent edit raw footage by conversation: it cuts from a word-level transcript, grades color, adds overlay animations and subtitles, then checks its own render.

    • Editing
    • Audio
    27.9kGitHub starsUpdated Oct 2026
  • JianYing Editor Skill

    luoluoluo22/jianying-editor-skill

    Lets a coding agent script the desktop JianYing Pro app (CapCut's China edition): import media, add subtitles, TTS voiceover, music, effects and keyframes into an editable draft.

    • Editing
    • Audio
    3.7kGitHub starsUpdated Sep 2026
  • OfficialPack

    VectCutAPI Skills

    sun-guannan/VectCutAPI

    Two skills for agents that build editing drafts: one drives the VectCutAPI service for CapCut and Jianying, the other covers CapCut desktop draft files, vertical captions and audio mixing.

    • Editing
    • Audio
    2.3kGitHub starsUpdated Sep 2026
  • OfficialPack

    OpenChatCut skills

    0xsline/OpenChatCut

    Four workflows bundled with the OpenChatCut video editor: AI short-film planning, shot-by-shot breakdown, beat-synced cutting and video model selection, loaded by its own agent or over MCP.

    • Editing
    • Storyboard
    • Generation
    2.1kGitHub starsUpdated Oct 2026
  • ffmpeg-skill

    kajisho5/ffmpeg-skill

    A local FFmpeg editing skill with 42 scripted tools for cutting, captioning, reframing, grading, audio clean-up and platform exports, plus an MCP interface.

    • Editing
    • Audio
    1.5kGitHub starsUpdated Oct 2026

This is a third-party skill. It is not maintained or audited by Pencil0, and skills can run scripts on your machine, so read the SKILL.md and any scripts before installing. Stars, license and dates are a GitHub snapshot and may be out of date.