Skills

Davinci Resolve Editor

Safe DaVinci Resolve Studio editing through davinci-resolve-mcp or the official Resolve...

Install

npx skills add liuluhaixiu/DaVinci-AutoEdit-Agent --skill davinci-resolve-editor -g -a claude-code -y
Category
AI and agents
Updated
5mo ago
License
MIT
Context cost
Lightabout 748 tokens
Runs scripts
No, instructions only

What it does

Safe DaVinci Resolve Studio editing through davinci-resolve-mcp or the official Resolve Python API. Use when Codex must inspect Resolve state, create or revise projects and timelines, import approved media, place source ranges, apply authorized LUTs or effects, prepare rendering, and verify gaps, tracks, source bounds, frame origin, and actual write success.

DaVinci Resolve Editor

Operate only from an approved blueprint. Read references/mcp-operations.md for the tool map and references/verification.md before delivery.

For deeper operations, load only the relevant bundled reference:

  • references/compound-tool-reference.md
  • references/prompt-patterns.md
  • references/page-capabilities.md
  • references/keyboard-shortcuts.md

Connect

  1. Confirm Resolve Studio is open and external scripting is set to Local.
  2. Call resolve -> get_version.
  3. If the tool is unavailable, distinguish missing registration from a server failure. An MCP config file alone is not proof of connection.
  4. Offer blueprint-only delivery or the official Python API fallback.

Install the upstream MCP server when the user approves configuration:

npx davinci-resolve-mcp setup

Write Contract

Before changing Resolve, state:

  • project and timeline name
  • tracks to create or modify
  • media and derivative paths
  • color, audio, subtitle, and render operations

Obtain explicit approval. Use a new project or duplicated timeline for substantial revisions. Preserve existing user tracks, audio, grades, and manual work unless the user explicitly authorizes modification.

Build

  • Import only blueprint media.
  • Convert seconds to frames using the actual project FPS.
  • Set source in/out and record frame deliberately.
  • Keep source ranges within probed bounds.
  • Use explicit track roles rather than assuming A1/A2/A3 universally.
  • Add BGM, LUTs, stabilization, retiming, audio processing, and renders only when the current request authorizes them.
  • Apply LUTs to verified camera/color-space groups and read back the node LUT.

Resolve API Boundaries

Read the whole SKILL.md on GitHub

More in AI and agents