Davinci MCP Professional

DaVinci MCP Professional

An enterprise-grade Model Context Protocol (MCP) server that exposes the full DaVinci Resolve scripting API to AI assistants. This project was inspired by the davinci-resolve-mcp project by @samuelgursky, even though it is now a significantly different architecture.

Compatible with any AI system that supports modern MCP specifications.

---

How It Works

Most MCP servers dump every available tool into the AI's context window at session start. With a large API surface, that means hundreds of tool definitions loaded whether or not the current task needs them — wasting tokens and slowing response time.

This server uses a kernel + domain architecture instead:

• Kernel tools (always available, 6 total): activate_domain, deactivate_domain, list_domains, get_version, get_current_page, switch_page. These are useful regardless of what you're doing in Resolve.
• Domain tools (loaded on demand): the full Resolve API is organized into domains — Project Management, Timeline Operations, Media Pool, and more. Call activate_domain("project_management") and the server fires a notifications/tools/list_changed event; your MCP client refetches the tool list and the domain's tools are immediately available, no session restart needed.

Context cost at session start is bounded by the kernel alone. You expand it deliberately, only for the domains your current workflow actually needs.

---

Who This Is For

This server is built for independent creators and boutique studios where AI API costs are real and context overhead matters.

Compared to other popular MCP servers for DaVinci Resolve:

Most alternatives load every available tool definition into the AI's context at session start. For a well-resourced production environment — a studio running Claude on dedicated infrastructure with token budgets absorbed into overhead — that approach is solid and the dense tool coverage is immediately useful.

This server makes a different architectural choice: 6 kernel tools at session start, domains activated on demand, context cost proportional to the work at hand. The full API surface (289 tools across 9 domains) is available, but none of it loads until you ask for it.

	Typical MCP server	davinci-mcp-professional
Tools at session start	All at once	6
Total tools available	Fixed	289 (9 domains)
Domain activation	—	On demand

If you're an independent creator or a small shop where every Claude API call has a cost, this architecture was designed for you. Professional-grade Resolve integration without professional-budget infrastructure requirements.

---

Prerequisites

• DaVinci Resolve Studio installed and licensed (the free edition does not support external scripting)

• Python 3.10 or later, installed system-wide via the official installer with "Add to PATH" and "Install for all users" selected

  Windows — critical: DaVinci Resolve locates Python through the Windows registry and loads python3.dll by full path from that installation. A uv-managed or user-only Python install uses a different DLL and will cause a two-runtime crash at connection time. Always create the virtual environment from the system Python (see Installation below).

• uv — fast Python package and virtual environment manager

---

Installation

git clone https://github.com/hoyt-harness/davinci-mcp-professional.git
cd davinci-mcp-professional

Windows — find your system Python path, then create the venv from it:

py -0p   # lists installed Python versions and their paths

uv venv --python "C:\Program Files\Python314\python.exe"   # adjust to your path
uv sync

macOS / Linux — uv venv with no --python flag works if the default python3 is a system-wide installation:

uv venv
uv sync

Configuring Claude Desktop

Locate or create claude_desktop_config.json:

• Windows: %APPDATA%\Claude\claude_desktop_config.json
• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

From source (Windows):

{
  "mcpServers": {
    "davinci-resolve": {
      "name": "DaVinci MCP Professional",
      "command": "C:\\path\\to\\davinci-mcp-professional\\.venv\\Scripts\\python.exe",
      "args": ["C:\\path\\to\\davinci-mcp-professional\\mcp_server.py"]
    }
  }
}

From source (macOS):

{
  "mcpServers": {
    "davinci-resolve": {
      "name": "DaVinci MCP Professional",
      "command": "/path/to/davinci-mcp-professional/.venv/bin/python",
      "args": ["/path/to/davinci-mcp-professional/mcp_server.py"]
    }
  }
}

Restart Claude Desktop after saving the config.

---

Configuring Claude Code

claude mcp add -s user davinci-resolve \
  -- /path/to/davinci-mcp-professional/.venv/Scripts/python.exe \
  /path/to/davinci-mcp-professional/mcp_server.py

This writes to ~/.claude.json and makes the server available in all Claude Code sessions without per-project configuration.

---

Basic Usage

1. Start DaVinci Resolve and wait for it to fully load.
2. Start the MCP server (Claude Desktop / Claude Code does this automatically).
3. The server starts with 6 kernel tools available. Use list_domains to see what domains are registered, then activate what you need:

What version of DaVinci Resolve is running?          # get_version — no activation needed
List the available domains.                           # list_domains
Activate the project management domain.              # activate_domain("project_management")
List all projects in the database.                    # list_projects — now available
Create a new timeline called "Act 1".                 # needs timeline_operations domain
Switch to the Color page.                             # switch_page — always in kernel

Domain overview

Domain	Activates	Tools
project_management	activate_domain("project_management")	Open/save/close/rename/delete projects, folder navigation, database switching
timeline_operations	activate_domain("timeline_operations")	Tracks, markers, timecode, export/import, generators, Fusion clips
media_pool	activate_domain("media_pool")	Folder management, clip operations, relink, mattes, stereo

Destructive operations require explicit confirmation

Any tool that permanently modifies or deletes data requires confirm: true in its arguments. If you call a destructive tool without it, the server returns a detailed error message describing exactly what would be destroyed and what you need to set to proceed. This applies to operations like delete_project, delete_timeline, delete_track, delete_media_pool_clips, and others.

---

Further Reading

Document	Purpose
USING.md	Developer setup, architecture, build instructions, contributing
BUGS.md	Troubleshooting and bug reporting
CONTRIBUTING.md	Contribution guidelines
COPYING	GPL-3.0 license