Your AI assistant just learned to color-grade

What it does

This is a Model Context Protocol (MCP) server that sits between AI assistants and DaVinci Resolve Studio. It translates natural-language requests into Resolve API calls, covering the full post-production pipeline: ingest, editing, color, Fusion, Fairlight, delivery, and even extension authoring. The server runs locally as a stdio process and ships with a browser-based control panel for inspecting Resolve state and reviewing analysis output.

The interesting bit

The project treats source media as immutable by default — analysis writes to sidecar directories, not your camera originals. It also exposes a clever dual-mode architecture: 32 compound tools for normal assistant use (grouped operations that stay within context limits), or 329 granular tools if you want one-to-one API method access. The analyze_media workflow can even hand frame paths back to the host AI for visual analysis, then commit those results into Resolve clip metadata.

Key highlights

• 100% API coverage (336/336 methods) with 98.5% live-tested against Resolve 19.1.3 and 20.3.2 Studio
• Source-safe media analysis: FFprobe/FFmpeg/Whisper workflows that never touch originals without explicit request
• 136 guarded workflow actions across 9 compound tools, plus a full granular mode
• Built-in control panel for local state inspection and inline analysis editing
• Auto-configures for Claude Desktop, Cursor, VS Code, Windsurf, Zed, Cline, Roo Code, JetBrains, and others
• Extension authoring helpers for DCTL, Fuse, and Lua/Python scripts with safe install/remove lifecycle

Caveats

• Requires DaVinci Resolve Studio (paid); the free edition blocks external scripting entirely
• Python 3.13+ compatibility depends on Resolve build version; 3.10–3.12 is the safest range
• Resolve 21 beta APIs are intentionally deferred until stable

Verdict

Post-production engineers and colorists who want AI assistance without surrendering their source media should look here. If you’re on the free Resolve tier or allergic to Python virtualenvs, this is a hard stop.