Workflow

An MCP Server Inside the Editor: What Resolve 21.1 Lets an Assistant Touch

ยท RenderBob team

DaVinci Resolve Studio 21.1 includes a native MCP server. Claude, Claude Code, and ChatGPT Codex can organise media, cut highlights, and batch render.

A guarded assistant core organizes media, proposes highlight cuts, and fills a render queue under explicit human permissions.

DaVinci Resolve Studio 21.1 includes a native MCP server. Claude, Claude Code, and ChatGPT Codex can organise media, cut highlights, and batch render. Scope the access before you hand it over.

Blackmagic released DaVinci Resolve 21.1 on 8 September. For pipeline work, the headline is a native Model Context Protocol server, so a supported assistant can call functions inside the application instead of only describing what to do.

What is supported

Blackmagic names Claude, Claude Code, and ChatGPT Codex. Setup runs through File, then Setup AI Assistants, and Resolve detects compatible assistants and configures the server. The tasks Blackmagic lists are analysing projects, organising media, adjusting settings, and batch rendering, with examples such as creating highlight edits from long-form video, removing unwanted clips, and rendering deliverables. The integration is Studio-only. Studio costs $295. CineD reported that the same release made advanced Python scripting Studio-only and dropped Python 2 support. Digital Production's write-up of the same release says the scripting API remains available from the console in the free edition, while workflow integrations and external scripting require Studio. Check the current feature list before you assume a free seat can drive a pipeline.

Where it fits in Resolve's year

Resolve 21 arrived in June with AI IntelliSearch, CineFocus, SlateFinder and SlateID, and the Photo page. IntelliSearch finds footage by description and SlateID reads slates into metadata. MCP lets a general assistant drive the project, which is a different layer from those tools. Palmier Pro, covered here on 23 September, took the same approach with a local MCP server in a generative timeline editor.

What to hand over

Mechanical, reversible tasks suit an assistant: sorting media into bins, logging, proposing a highlight reel as a new timeline, batch renders. Our earlier two-lanes piece drew the line between cleanup that follows decisions and assembly that makes them. The same line applies. A tutorial on the integration notes that transcript accuracy decides whether an assistant's cut is usable, so correct the transcript first.

Controls to set before the first prompt

  • Work on a duplicate project, and keep a version you can return to.
  • Ask for a new timeline instead of changes to the locked one.
  • Treat deletion and render-over requests as ones that need your confirmation.
  • Check what changed after each run, not only what you asked for.
  • Keep client footage in mind: an assistant that can read a project can read everything in it, so check where the assistant's own processing runs.

What to watch

Whether Adobe and Apple add similar protocol support, whether the exposed tool set grows beyond media management, and whether projects carry a record of which edits an assistant made.

More from the blog

All posts