> ## Documentation Index
> Fetch the complete documentation index at: https://aspect.inc/docs/llms.txt
> Use this file to discover all available pages before exploring further.
# Connect DaVinci Resolve
> Set up Blackmagic's official Resolve MCP with Claude, Codex, or ChatGPT desktop, then pair it with Aspect.
Blackmagic Design includes a native MCP server with **DaVinci Resolve Studio 21.1**. It runs on the computer with Resolve and exposes the application's scripting API to an AI assistant. This guide covers that bundled server, rather than a separately maintained community server.
Blackmagic documents **File → Setup AI Assistants** in the [Resolve 21.1 reference manual](https://documents.blackmagicdesign.com/UserManuals/DaVinci_Resolve_21.1_Reference_Manual.pdf), in “Using DaVinci Resolve with AI Assistants.” Download Studio through [Blackmagic's support center](https://www.blackmagicdesign.com/support/family/davinci-resolve-and-fusion).
## Connect from the File menu
Install DaVinci Resolve **Studio 21.1 or later** and the desktop assistant you intend to use on the same computer. You can connect them from Resolve without opening a terminal.
Open Resolve, click **File** in the top menu, and choose **Setup AI Assistants…** near the bottom of the dropdown.
Read the **AI Assistants Configured** confirmation to see which assistants Resolve found. The list depends on what is installed on your computer. This example includes **Codex in ChatGPT**.
If **Claude Desktop** opens an extension installer, finish that installation too.
Click **OK**, then quit and reopen the assistant you want to use. Open the intended Resolve project and ask: “Tell me the name of my open Resolve project and how many timelines it contains. Do not change anything yet.”
Confirm that it identifies the right project before asking for an edit.
The sections below cover individual client settings and [terminal setup](#terminal-setup-for-codex-and-claude-code). Use them if your assistant is not configured by the File menu option, or when managing a connection manually.
## Local scripting settings
For a local scripting connection, check **Preferences → System → General → External scripting** and use **Local** when required. You do not need to open network scripting for an assistant running on the same computer. Keep **Automatic scripted actions** at **Allow safe** unless a specific workflow requires something else.
The available setup targets depend on the installed Resolve and client versions.
## Claude Desktop
Use Resolve's setup command first. If you need to install the bundled extension manually, open **Claude → Settings → Extensions → Advanced settings → Install Extension** and select Blackmagic's `DaVinciResolve.mcpb` file.
On a standard macOS installation, it is at:
```text theme={null}
/Applications/DaVinci Resolve/DaVinci Resolve.app/Contents/Resources/DaVinciResolve.mcpb
```
This package is authored by Blackmagic Design and starts the `ResolveMCP` executable from the installed Studio application. It uses Claude Desktop's extension runtime; it does not require installing a third-party Python MCP project. See [Claude's local-extension instructions](https://support.claude.com/en/articles/10949351-getting-started-with-local-mcp-servers-on-claude-desktop).
## Terminal setup for Codex and Claude Code
The native executable speaks MCP over **standard input and output**, usually called **STDIO**. Configure it as a local command with no server arguments.
| Operating system | Default executable |
| - | - |
| macOS | `/Applications/DaVinci Resolve/DaVinci Resolve.app/Contents/Applications/ResolveMCP` |
| Windows | `C:\Program Files\Blackmagic Design\DaVinci Resolve\ResolveMCP.exe` |
| Linux | `/opt/resolve/bin/ResolveMCP` |
Check the actual installation path, especially on Windows if you chose another drive. These are filesystem paths, not HTTP server URLs.
For Codex on macOS:
```bash theme={null}
codex mcp add davinci_resolve -- "/Applications/DaVinci Resolve/DaVinci Resolve.app/Contents/Applications/ResolveMCP"
codex mcp list
```
For Claude Code on macOS:
```bash theme={null}
claude mcp add --transport stdio --scope user davinci_resolve -- "/Applications/DaVinci Resolve/DaVinci Resolve.app/Contents/Applications/ResolveMCP"
claude mcp list
```
Use the matching executable path for your operating system. If the connection already exists, inspect or edit it rather than adding a duplicate. Start a fresh session after setup. The native binary uses Resolve's bundled runtime; a separate Python installation and legacy `PYTHONPATH` configuration are not needed for this connection.
See the current [Codex MCP configuration](https://learn.chatgpt.com/docs/extend/mcp) and [Claude Code MCP documentation](https://code.claude.com/docs/en/mcp).
## ChatGPT desktop
In the ChatGPT desktop app, open **Settings → MCP servers → Add server**. Choose **STDIO**, name the connection `davinci_resolve`, and enter the full executable path from the table above. Leave arguments empty, save, and restart the connection.
Use a conversation that has access to the local computer and its MCP tools. Confirm that `get_resolve_status` is available before asking for an edit. [OpenAI's MCP guide](https://learn.chatgpt.com/docs/extend/mcp) describes the shared local-host configuration and current surface support.
ChatGPT web does not read this computer's MCP configuration. A cloud-only chat cannot launch a local `ResolveMCP` executable from its path. Use a supported desktop/local-computer workflow for Resolve, or an explicitly configured private connection for your environment. Aspect's remote MCP and Resolve's local MCP are separate connections.
## Verify the connection before editing
Start with this request:
```text theme={null}
Use the Resolve MCP to report whether Resolve is running and its version.
Read the available scripting API for project and timeline inspection.
Tell me whether a project is open and how many timelines it contains.
Do not create or change a project yet.
```
The server provides `get_resolve_status`, `get_whats_new`, `search_scripting_api`, `get_scripting_api`, and `get_scripting_docs`. If Resolve is closed, `launch_resolve` can start it. Let the application finish opening before inspecting a project.
For an agent using `run_script`, this is a read-only example after discovering the relevant API:
```python theme={null}
result = {
"version": resolve.GetVersionString(),
"project_open": project is not None,
"timeline_count": project.GetTimelineCount() if project else 0,
}
```
`resolve` and `project` are provided by the server. Assign `result` to return structured data. This example was checked with the official server on macOS and Resolve Studio **21.1.0.17**. That check confirms the connection and read-only scripting, rather than every editing operation on every platform.
## How the agent performs an edit
The official server exposes a small set of general tools. Most editing work goes through `run_script` after the agent discovers the current scripting API. It does not need a separate MCP tool for every timeline operation.
For example, the API includes `MediaPool.AppendToTimeline`, `MediaPool.RelinkClips`, and `MediaPoolItem.LinkProxyMedia`. Have the agent inspect the current signatures before writing a script. Some interface actions may not have a scripting equivalent.
Ask for a named working timeline and specific changes, then inspect the result in Resolve. Start with a short assembly or a few clips before applying a long batch of edits. A successful script call should be followed by a check that the intended clips, ranges, and tracks actually changed.
`run_script` has access to Resolve but blocks general filesystem, network, and process operations. It can still change a project. `run_script_unsafe` adds system access and is unnecessary for ordinary API-only inspection or timeline operations. Keep file transfers in the Aspect CLI and use the narrower script tool when it can do the job.
## Pair Resolve with Aspect
Connect [Aspect MCP](/docs/mcp/index) to the same assistant and install the [Aspect CLI](/docs/cli/installation). The two servers handle different parts of the work:
1. Aspect MCP finds relevant footage, transcript passages, scene descriptions, and review notes.
2. Aspect supplies proxy files and originals through the mounted project. The CLI starts pins and reports their state.
3. Resolve MCP turns those local media paths and selected ranges into editing operations in Resolve.
4. After pinning completes, Resolve uses the originals for finishing. Aspect can hold the review render and its feedback.
Follow [Agentic Video Editing](/docs/workflows/agentic-video-editing) for the complete folder-copy, proxy, pinning, and relink workflow.
```text theme={null}
Use Aspect MCP to find the interview selects for this brief and record
source paths and time ranges. Read the Aspect workflow for Resolve.
Prepare compatible proxies on the mount and start pinning the originals.
In Resolve, create a new working timeline and assemble a short first pass.
Check proxy matches and report any offline clips. Before final delivery,
verify original pins are synced and confirm the timeline uses originals.
```
## Common setup problems
* **Setup AI Assistants is missing:** check that this is the Studio edition and a version with the bundled MCP server.
* **The server starts but Resolve cannot be reached:** open Resolve fully, check local scripting access, and restart the connection.
* **Tools are absent after installation:** restart the AI client and inspect its server list. Check whether workspace settings allow local extensions or MCP servers.
* **The agent uses old scripting methods:** ask it to read the installed API through `search_scripting_api` or `get_scripting_api` before retrying.
* **Media is offline:** confirm the mount path exists on the editing computer. An `aspect://` address identifies a cloud location; Resolve imports the corresponding filesystem path from the mount.
This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.