Start Here: Open 00_START_HERE.html

For the clearest setup, open 00_START_HERE.html first. It is the visual, step-by-step guide for macOS, Windows, Antigravity, Claude Code, Codex, Cursor, and Remotion. Keep this entire folder together while using it.

Resolve AI Bridge

Resolve AI Bridge is an MIT-licensed local MCP bridge that lets a compatible AI client inspect and control the project currently open in DaVinci Resolve.

The design is intentionally small:

1. ResolveConsole.py starts inside Resolve's Python 3 Console and captures the internal Resolve API object.

2. A daemon thread reads authenticated JSON jobs from ~/.resolve-ai-bridge/inbox while the Console remains available.

3. bridge/server.py exposes a focused set of MCP tools to your AI client over stdio.

4. Replies return through ~/.resolve-ai-bridge/outbox.

There is no hosted bridge account and no network listener. Your AI provider may still process prompts according to its own privacy policy.

Requirements

• DaVinci Resolve with the internal Python 3 Console available

• Python 3.10 or newer on the computer

• An MCP-compatible client such as Antigravity, Claude Code, Codex, or Cursor

• Internet access during installation so pip can install the Python MCP package

• Node.js current LTS, heavily recommended for Remotion motion graphics

The bridge itself does not require Node.js. Node.js is for the recommended Remotion workflow.

Quick Setup

1. Download the complete repository

On GitHub select Code, then Download ZIP, and extract it. Do not copy only install.py; it needs the agent, bridge, and skills folders beside it.

You may rename the downloaded repository folder. The installer copies runtime files to a fixed location so later repository renames do not break the Console line.

2. Run the installer

macOS:

python3 install.py

Windows PowerShell:

py install.py

The installer:

• Copies the runtime to ~/.resolve-ai-bridge

• Preserves or creates a private rab_... token

• Creates ~/.resolve-ai-bridge/.venv

• Installs the official Python MCP dependency into that venv

• Writes filled MCP JSON plus exact Console, Claude Code, and Codex command files

• Installs the included Resolve editing skill for Claude Code and Codex skill locations

It does not rewrite an AI client's existing MCP settings.

3. Start the agent inside Resolve

1. Open Resolve and the project you want to edit.

2. Open Workspace > Console.

3. Select Py3.

4. Paste the exact line printed by the installer.

macOS shape:

exec(open("/Users/YOUR_NAME/.resolve-ai-bridge/ResolveConsole.py", encoding="utf-8").read())

Windows shape:

exec(open(r"C:\Users\YOUR_NAME\.resolve-ai-bridge\ResolveConsole.py", encoding="utf-8").read())

The original installed filename is ResolveConsole.py. Do not rename that installed file. If it was moved or renamed, run the installer again and use its newly printed line.

The Console should immediately print:

• RESOLVE AI BRIDGE READY

• The local token

• A single-server MCP object

• A full mcpServers JSON block

The worker is a daemon thread, so the pasted line returns and releases the Console immediately.

To stop it in the same Resolve session:

__resolve_ai_bridge_runtime__.stop()

4. Add the MCP server to your AI client

Use the filled JSON printed by Resolve or open:

~/.resolve-ai-bridge/mcp-config.json

Its shape is:

{
  "mcpServers": {
    "resolve-ai-bridge": {
      "command": "/ABSOLUTE/PATH/TO/.resolve-ai-bridge/.venv/bin/python",
      "args": [
        "/ABSOLUTE/PATH/TO/.resolve-ai-bridge/bridge/server.py"
      ],
      "env": {
        "RESOLVE_AI_BRIDGE_TOKEN": "RAB_TOKEN_PRINTED_IN_RESOLVE"
      }
    }
  }
}

Do not use ~ or a relative path in MCP settings. Use the absolute paths generated by the installer.

Antigravity

1. Open the Agent panel.

2. Open the three-dot menu and select MCP Servers.

3. Choose Manage MCP Servers, then View raw config.

4. Merge the resolve-ai-bridge entry into the existing mcpServers object.

5. Preserve every server already in the file.

6. Save, refresh the MCP list, and restart Antigravity if needed.

Opening raw config through the application avoids guessing which config path a particular Antigravity version uses.

Claude Code

The installer and Resolve Console print an exact command that safely reads the generated JSON file. Run that printed command in a normal terminal. It has this macOS shape:

claude mcp add-json resolve-ai-bridge "$(cat ~/.resolve-ai-bridge/claude-server-entry.json)" --scope user

On Windows, use the PowerShell command printed by the installer. The exact command is also saved at ~/.resolve-ai-bridge/claude-command.txt. Restart Claude Code and run /mcp. If your installed version does not support add-json, merge the full JSON through its MCP settings instead.

Codex

The Resolve Console and installer both print an exact command for your installed paths and token. It has this shape:

codex mcp add resolve-ai-bridge --env RESOLVE_AI_BRIDGE_TOKEN=TOKEN -- /absolute/path/to/.venv/python /absolute/path/to/bridge/server.py

Run the filled command in a normal terminal, then use codex mcp list. The same filled line is saved at ~/.resolve-ai-bridge/codex-command.txt. The Codex app and CLI also let you add a local STDIO server through Settings > MCP servers.

Cursor

Open Cursor Settings, search for MCP, choose Add new global MCP server, and merge the generated entry into mcpServers. Save and enable it.

5. Verify before editing

Keep the Console agent running, then run:

python3 tools/doctor.py

On Windows:

py tools/doctor.py

Then ask your AI:

Call resolve_status and tell me the open project and timeline. Do not edit anything.

For a safe first write:

Inspect the timeline. Add a blue marker at the current playhead named Bridge test, then inspect again to verify it.

Token Behavior

The token is functional, not decorative.

• The installer writes it to ~/.resolve-ai-bridge/token.txt with restricted permissions where supported.

• The MCP process receives the same token through its environment.

• Every job includes the token.

• The agent compares it before executing an operation.

• A mismatched request is rejected.

The token protects this local queue from accidental or unrelated local requests. It is not an internet login. Do not commit it, post it, or paste it into a public issue.

Rotate it if exposed:

python3 install.py --rotate-token

Then restart the Resolve agent and replace the old MCP entry in your AI client.

MCP Tools

The server intentionally exposes fewer than 20 focused tools so a model can choose them reliably.

The MCP server also provides resolve://guide and an edit_video prompt.

Getting Better AI Edits

The included resolve-ai-editing skill uses an inspect, plan, edit, verify workflow. The installer copies it to Claude Code and Codex skill folders and keeps a reference copy at:

~/.resolve-ai-bridge/skills/resolve-ai-editing/SKILL.md

Start complicated requests with:

First inspect the open project and timeline. Summarize what is there, state a short edit plan, and list assumptions. Work in small steps and verify after each change. Ask before deleting clips or starting a render.

Good prompts include:

• Target platform and aspect ratio

• Desired duration

• Audience and story goal

• Reference style

• What must not change

• Whether deletion and rendering are allowed

Skills improve API usage and planning. They do not make subjective editing quality automatic. Review the timeline and use project backups.

Remotion Setup, Heavily Recommended

Remotion creates video with React. It is useful for motion graphics, animated type, explainers, and designed scenes that are awkward through Resolve's scripting API.

Remotion does not control Resolve and is not installed inside the Python bridge. The practical workflow is:

1. Build and preview a designed clip with a coding agent and Remotion.

2. Render the clip to a local media file.

3. Import that file through Resolve AI Bridge.

4. Ask where it should be placed before appending it.

Install Node.js

Install the current LTS release from nodejs.org. Open a new normal terminal and verify:

node --version
npm --version

Install the official Remotion skill and create a project

Run in a normal terminal, not the Resolve Console:

npx -y skills@latest add remotion-dev/skills -g -y
npx create-video@latest --yes --blank my-video
cd my-video
npm install
npx remotion skills add
npm run dev

The maintained remotion-dev/skills bundle teaches supported coding agents Remotion composition, timing, media, and animation practices. Open your AI client in the my-video folder.

Example prompt:

Use the official Remotion skill. Before coding, propose a 15-second 1920x1080 motion graphic with three scenes. Use my exact copy, restrained typography, and smooth transitions. Keep important content inside safe margins. Build it, run checks, and open Remotion Studio so I can review it before rendering.

Render the approved composition:

npx remotion render <CompositionId> out/video.mp4

Then give the absolute output path to your Resolve AI:

Import /absolute/path/to/my-video/out/video.mp4. Inspect the current timeline and ask me where to place it before appending it.

See docs/REMOTION.md for a detailed workflow and prompt checklist.

Architecture

AI client
  |  MCP over stdio
  v
~/.resolve-ai-bridge/bridge/server.py
  |  authenticated JSON files
  v
~/.resolve-ai-bridge/inbox and outbox
  |
  v
ResolveConsole.py daemon thread inside Resolve
  |
  v
Current Resolve project and timeline

Known Boundaries

• Resolve's public scripting API does not expose every action from the Edit page.

• Transition creation and arbitrary timeline repositioning are not offered because they cannot be implemented reliably with the documented API surface.

• Background-thread access to Resolve's internal objects can vary by Resolve build. The Console returns immediately by design, but real compatibility must be verified on your exact version.

• Render presets must already exist in Resolve.

• Complex Fusion automation is intentionally not included in this first simple version. Use Remotion for designed motion, or add Fusion operations only after testing them on real projects.

• The bridge works on the currently open local project. It is not a remote collaboration server.

Verification Status

• The React instructions application is built with Vite as part of repository verification.

• tools/doctor.py compiles every Python source in memory before checking the installation.

• A real Resolve session is still required to verify the injected API object and each editing operation on a specific Resolve build.

Do not interpret a successful web build as proof that a particular Resolve API operation works on every Resolve version.

Repository Layout

00_START_HERE.html             visual setup guide
README.md                      complete written reference
install.py                     cross-platform installer
install-macos.command          macOS double-click wrapper
install-windows.bat            Windows double-click wrapper
requirements.txt               Python MCP dependency

agent/ResolveConsole.py        internal Resolve Console agent
bridge/client.py               authenticated file queue
bridge/server.py               MCP tools over stdio
tools/doctor.py                syntax, install, heartbeat, and round-trip checks
skills/resolve-ai-editing/     included Agent Skill
docs/                          troubleshooting, recipes, and Remotion guide
src/                           source for the visual guide

Development

Web guide:

npm install
npm run dev
npm run build

Python syntax and local installation checks:

python3 tools/doctor.py

Important contributor rules are in AGENTS.md.

Need Help From Your Provider?

After following the steps above, paste this into Antigravity, Claude Code, Codex, or another local coding agent:

Read 00_START_HERE.html and README.md in this project. Run tools/doctor.py, inspect ~/.resolve-ai-bridge/mcp-config.json, and configure your own MCP settings for resolve-ai-bridge. Preserve my existing MCP servers. Do not change or expose the token. Test with resolve_status only and explain any error in simple steps.

License

MIT. See LICENSE.