Skip to content
AI Knowledge MapSearch / Text directory

Chapter 2 Use an agent to operate professional video editors

Reading note

After generating shots, you generally still need to adjust clip lengths, shot order, composition, color and sound. This chapter first chooses professional editors, then connects an agent so you can operate an existing project through natural language and refine it after previewing.

1. Choose between two editors

This chapter selects DaVinci Resolve Studio and Adobe Premiere Pro for their professional postproduction capabilities, editable projects and documented agent connections. Choose either one to learn.

PlatformWhy it was selectedAgent connection routeHow to choose
DaVinci Resolve StudioEditing, grading, Fusion compositing and Fairlight audio in one projectVersion 21.1 added official AI assistant integration; third-party MCP is another option belowStart here for a new postproduction workflow emphasizing integrated image and audio work
Adobe Premiere ProPrecise timeline editing, text and graphics, and Adobe asset collaborationBuilt-in AI Assistant (beta); third-party PPMCP for an external agentContinue with it if you already use Photoshop, After Effects or Premiere projects

An editor having a feature and an agent being able to call it are separate questions. Begin by connecting queries, one small edit and project saving, then expand the task.

2. DaVinci Resolve Studio: connect an external agent

Official integration and the route used here

Blackmagic confirmed AI assistant integration in its official Resolve 21.1 release notes. Listed tasks include project analysis, media organization, setting changes, highlight edits and batch rendering. For that version, consult the 21.1 manual and new-feature guides in the official support center and use the connection method supplied with the software.

The instructions below cover the third-party samuelgursky/davinci-resolve-mcp route, whose installation documentation provides environment preparation, client configuration and connection checks. This and the official integration are separate solutions; choose one. Project documentation

1. Prepare the editor and scripting environment

The author requires Resolve Studio 18.5 or later, Git, and 64-bit Python 3.10 or later; the commands below use Python 3.12. Media analysis also needs ffmpeg available on PATH; add dependencies for transcription and other features as needed. Start Resolve and open a practice project. Set Preferences → System → General → External scripting using to Local, then save. Installation and version requirements

This section is written for Studio. Scripting and bridge conditions in the free edition vary by version; do not apply these steps unchanged.

2. Download the project and generate client configuration

Open PowerShell in your chosen tools directory and run:

git clone https://github.com/samuelgursky/davinci-resolve-mcp.git
Set-Location davinci-resolve-mcp
py -3.12 install.py --clients manual

The installer creates the runtime environment and prints manual connection settings. Add its generated command, arguments and environment variables to your agent client using the actual paths it supplies. Merge the server entry if other MCP configurations exist. The same author's installation guide covers other installation methods and troubleshooting.

3. Connect and inspect the current project

Your client must support launching a local stdio MCP service. Save the configuration, restart the client and enable the Resolve tools while leaving Resolve and the project open. First ask the agent to read the project name, timelines, tracks and clips, then compare its response with the editor.

4. Begin with explicitly supported edits

Start with media organization, clip-property changes, markers or render settings. Before cutting, moving or replacing clips, ask the agent to check that its interface supports the operation. The timeline operation limits distinguish supported, partially supported and restricted operations; an interface may not expose everything available in the editor UI.

Editor and connection tutorials: official Blackmagic training, MCP installation guide, timeline capabilities.

3. Adobe Premiere Pro: built-in assistant and external agents

Route A: built-in AI Assistant

If your version and account include the assistant, open a project in Premiere and choose Window → Assistant. Adobe's official overview still labels it beta; listed tasks include bins and renaming, transcription, markers and initial clip assembly.

Use this route to organize media and build a rough cut with natural language. For precise clip, composition or volume edits, check the assistant's current supported actions, or choose the external route below.

Route B: external agent + PPMCP

This chapter uses the third-party project CaYatur/PremiereProMCP (PPMCP). A UXP plugin and local bridge connect Premiere and expose editing tools to an agent. The following instructions use Windows, the author's primary testing environment.

1. Prepare Premiere and the plugin loader

Use Premiere 26.0 or later and install Adobe UXP Developer Tool. Enable Enable developer mode in Premiere's Settings → Plugins, then restart Premiere. Adobe's general UXP tutorials and this project have different minimum versions; this section follows the project's 26.0 requirement. Project requirements, Adobe plugin loading documentation

2. Install PPMCP

Download PPMCP-Setup-x.x.x.zip from the author's releases. Extract it completely, keeping the relative locations of Setup.bat and payload, then run Setup.bat. After installation, open the generated instructions:

  • %APPDATA%/PPMCP/HOW-TO-USE.txt: usage and actual local paths.
  • %APPDATA%/PPMCP/HOW-TO-CONNECT.txt: application, server and plugin entry points.
  • %APPDATA%/PPMCP/mcp-config-snippet.json: configuration to merge into your client.

This is the author's release-package installation method; it has not been run on this site's local machine. Full installation guide

3. Start the bridge and load the plugin

Run the PPMCP Bridge shortcut created by the installer. Open Premiere and UXP Developer Tool; select Add Plugin in the latter using the plugin manifest path in HOW-TO-CONNECT.txt, then click Load. Open the PPMCP panel in Premiere and confirm Active status. Reopen a closed panel through Window → UXP Plugins. Author's connection steps, Adobe panel entry points

4. Configure the agent client

Merge the generated MCP snippet into the client, save and restart it. PPMCP uses local stdio: the client launches the specified Node program with the server entry file as an argument. It cannot be entered into a connector accepting only remote MCP URLs. Configuration formats vary; follow the author's client configuration section using the installer's actual paths.

5. Verify supported operations

The author's recorded real-machine checks include reading projects and sequences, some clip edits, position and scale, audio gain, project saving and sequence screenshots. The full tool list also includes features not individually tested; tool count alone does not establish reliability. Tools and validation status

Editable MOGRT text needs the author's additional CEP Text Bridge. A PNG text overlay is different from editable captions. Creating tracks and other operations can also be constrained by UXP; check the current version's support first.

Editor and connection tutorials: official Premiere help, built-in assistant overview, PPMCP installation guide, tool capabilities and limits.

4. Troubleshooting and result checks

SymptomCheck first
Agent gives advice but the editor does not changeActual editing tools are enabled, calls are recorded, and project readback identifies the right project
Resolve service starts but cannot read a projectStudio version, local scripting settings, running Resolve and an open project
Premiere plugin is not connectedDeveloper mode, loaded UXP plugin, running PPMCP Bridge and Active panel status
MCP tool list is emptyClient support for local stdio, valid launch paths and correctly merged client-specific configuration
A menu feature cannot be called by the agentCurrent interface/tool support; retain the error and complete that step in the editor
Saving succeeds but there is no test videoCheck project saving and video export separately, including render status and the actual file