samuelgursky/davinci-resolve-mcp

What this is

davinci-resolve-mcp is a bridge that lets AI assistants (like Claude) directly control DaVinci Resolve, a professional video editing and color grading application. It implements the Model Context Protocol (MCP) — a standard way for AI tools to connect to external software — so an AI can perform real editing tasks: opening projects, building timelines, adding clips, adjusting color, setting up renders, and more.

How it works

DaVinci Resolve ships with an official scripting API that lets external programs control it. This project wraps that API into MCP "tools" that an AI assistant can call. When you ask an AI "create a timeline called 'Assembly Cut' from all clips in the current bin," the AI translates that into a series of API calls through this server.

It ships in two modes:

• Compound mode (default): 37 tools that group related operations together, keeping AI context usage low
• Full/granular mode: 390 tools, one per Resolve API method, for power users

There's also an advanced offline server (Node.js) that works without Resolve running — it reads and edits Resolve project files (.drp, .drt, .drx) directly, enabling cloud-based grading, QC, and pipeline automation.

Key features

• Full editing control: media pool management, timeline editing, color grading, Fusion compositing, audio/Fairlight, render setup, review markers
• Media analysis: detects sync events (2-pops, slate claps), analyzes clips, writes metadata back into Resolve
• Safety envelope: every tool returns a structured result with status, verification (did the readback confirm the change?), and semantic delta — so AI agents know if an action actually succeeded
• Execution traces: multi-step AI operations are logged as unified traces, letting editors audit why the AI did something
• Free edition support: Blackmagic gates external scripting to the paid Studio version, but this project includes an in-app bridge script that works around that for the free edition (on Resolve 21.0.x)
• Local control panel: a browser-based UI to inspect Resolve state and analysis results
• Client support: auto-configures Claude Desktop, Cursor, VS Code, Windsurf, Zed, and many other AI coding tools

Why it matters

This is a practical example of AI agents operating professional creative software — not just chatting, but doing real work in a tool where mistakes are costly. The emphasis on verification, audit trails, and guarded operations (checking before destructive actions) shows how agentic AI can be made safe enough for production environments. The offline file-level server also demonstrates how AI pipelines can work in the cloud without a running desktop app.