> ## Documentation Index
> Fetch the complete documentation index at: https://docs.premierecopilot.com/llms.txt
> Use this file to discover all available pages before exploring further.

# MCP Connector (Claude, ChatGPT)

> Connect PremiereGPT to Claude, ChatGPT or any MCP client, then edit your Premiere Pro or DaVinci Resolve project from the chat.

The PremiereGPT MCP connector plugs PremiereGPT into the AI assistant you already use: **Claude**, **ChatGPT**, **Gemini**, **Claude Code** or any other client that supports MCP (Model Context Protocol). You ask in the assistant's chat, and it edits the project open in **Premiere Pro** or **DaVinci Resolve** through the PremiereGPT panel: reading the timeline, cutting silences, adding captions, building vertical clips, and more.

Your project never leaves your computer to be edited: the assistant sends instructions, the PremiereGPT panel runs them in your editor.

In assistant directories (Claude, ChatGPT), the connector is listed as **PremiereCopilot**. Its tools start with `premierecopilot_`; the older `premieregpt_` names still work.

## Requirements

* **Premiere Pro** or **DaVinci Resolve** open, with a project and a sequence.
* The **PremiereGPT panel** installed, open and signed in (in DaVinci Resolve the panel is called DavinciGPT). See [Installation](https://docs.premierecopilot.com/getting-started/installation.md).
* The **same PremiereGPT account** in the panel and in the assistant: the assistant can only reach panels signed in to the account it was connected with.
* **Credits** for the paid tools (transcription, Jump Cut, AI rough cut, Smart Virals analysis, captions…). Reading the timeline and many editing tools are free: see the [tool list](#tools).

> [!NOTE]
>
> Keep your editor and the panel open while the assistant works: every action goes through the panel.

## Connect your assistant

Server URL:

```
https://mcp.premierecopilot.com/mcp
```

Sign-in uses OAuth: a PremiereGPT page opens, you sign in with your PremiereGPT email and password and allow the assistant. You never paste a key or a token.

You can also find the URL and these steps in the panel: click the Claude, ChatGPT or Gemini logo in the top bar (or **Connect Claude, ChatGPT…** in the **⋯** menu) to open the **MCP** page. It turns live as soon as an assistant connects.

> [tab] Claude
>
> **1. Open the connectors**
>
> In Claude, open **Settings → Connectors** (shown as **Customize → Connectors** in some versions).
>
> **2. Add a custom connector**
>
> Click **+ Add**, then **Add custom connector**. Name it **PremiereCopilot**, paste the URL above, then click **Continue** and **Add**.
>
> **3. Sign in**
>
> Click **Connect**, sign in with your PremiereGPT account and click **Sign in and allow**. You are sent back to Claude.
>
> > [!NOTE]
> >
> > On the free plan, Claude allows one custom connector. On Team and Enterprise plans, an owner adds the connector in the admin settings.

> [tab] ChatGPT
>
> **1. Open the plugins**
>
> In ChatGPT on the web, open **Plugins**.
>
> **2. Add a custom MCP server**
>
> Click **+**, then **Add custom MCP server**. Name it **PremiereCopilot**, paste the URL above and choose **OAuth** as the authentication.
>
> **3. Sign in**
>
> Tick the confirmation box, click **Create as a plugin**, then sign in with your PremiereGPT account and allow ChatGPT.
>
> > [!NOTE]
> >
> > Editing tools need ChatGPT Business or Enterprise (an admin adds the server). On other plans, ChatGPT may only offer the read-only tools. Setup is done on the web.

> [tab] Gemini
>
> **1. Open the connected apps**
>
> On gemini.google.com, open **Settings → Connected Apps**.
>
> **2. Add a custom app**
>
> Under **Custom apps**, click **Add a custom app** and paste the URL above.
>
> **3. Sign in**
>
> Click **Next** and sign in with your PremiereGPT account.
>
> > [!NOTE]
> >
> > Custom apps in Gemini currently need a personal Google account, 18 or older, in the United States, with Gemini Apps Activity on. Add it on the web: it then works on mobile too.

> [tab] Claude Code and other clients
>
> Any client that supports remote MCP servers over **Streamable HTTP** with OAuth can connect.
>
> In Claude Code:
>
> ```bash theme={null}
> claude mcp add --transport http premierecopilot https://mcp.premierecopilot.com/mcp
> ```
>
> Then type `/mcp` in Claude Code, select **premierecopilot** and sign in with your PremiereGPT account in the browser page that opens.
>
> For another client, add a remote (HTTP) MCP server with the URL above. The client discovers the sign-in page on its own: no API key is needed.

> [!TIP]
>
> Ask *"Is PremiereGPT connected?"* to check the link: the assistant tells you whether the panel is open and in which editor.

## What you can ask

Talk to the assistant as you would to an editor. A few requests that work well:

> *"Remove the silences from my active sequence with a balanced rhythm."*

Jump Cut runs on the range you confirm. The sequence is backed up first.

> *"Transcribe the interview clip, then add karaoke captions to the whole timeline."*

The assistant asks before transcribing (it uses credits), then places the captions in the Karaoke style of Smart Captions.

> *"Find the 3 most viral moments of this podcast and create a vertical sequence for each."*

Smart Virals rates every sentence, the assistant picks the best hooks and payoffs, and PremiereGPT creates one vertical sequence per viral. Your source sequence does not change.

> *"Make a rough cut of this talking-head video: remove the retakes and the off-camera talk, aim for about 8 minutes."*

The AI rough cut tool reads the transcription, scores every sentence and removes what is not needed.

> *"Show me the timeline."*

A visual preview of your active sequence appears in the conversation.

> [!TIP]
>
> Be precise about the range ("from 2:10 to 5:00", "the whole timeline") and the result you want. The assistant reads the timeline before acting and confirms the destructive edits with you.

## Tools

The tools come in three families: `timeline_*` work on the active sequence, `project_*` on the project (bins, sequences, items), `premierecopilot_*` run PremiereGPT features. You do not need to name them: the assistant picks them from your request.

> [!NOTE]
>
> **Track names start at 0 in every connector tool.** `V0` and `A0` are the first video and audio tracks, shown as **V1** and **A1** in your editor. If the assistant mentions `A1`, it means the second audio track of your timeline.

### Timeline

| Tool | What it does | Changes your project | Uses credits |
| - | - | - | - |
| `timeline_get_active` | Name and id of the active sequence. | No | No |
| `timeline_read` | Reads the active sequence: format, tracks, clips, markers, transcribed speech. | No | No |
| `timeline_preview` | Shows a visual preview of the timeline in the conversation. | No | No |
| `timeline_see_frame` | Exports up to 8 frames so the assistant can look at the picture. | No | No |
| `timeline_open` | Opens a sequence. | No | No |
| `timeline_edit` | Deletes, trims, splits, moves, adds or disables clips, or a whole track at once (mute a duplicate camera sound); adds markers, transitions and tracks. | Yes | No |
| `timeline_create_edit` | Builds a whole new edit in a new sequence from project media. | Yes (new sequence) | No |
| `timeline_edit_effects` | Adds or edits video and audio effects, keyframes and clip speed, on chosen clips or every clip of a track. | Yes | No |
| `timeline_transform` | Positions, scales and crops clips (reframe, split screen, picture-in-picture at an exact size and margin), on chosen clips or every clip of a track. | Yes | No |
| `timeline_round_corners` | Rounds the corners of clips where the picture ends on screen, even when the clip is zoomed or moved (Premiere's Film Impact rounded crop), on chosen clips or every clip of a track. | Yes | No |
| `premierecopilot_design_image` | Renders a still design written by the assistant (background, frame, shadow, band) as a transparent image and places it on the track you choose, even under your video. | Yes | No |
| `timeline_reframe` | Reframes footage to the sequence format, keeping the subject in frame. | Yes | Yes |
| `timeline_read_effects` | Reads the real effects and settings of clips (position, scale, crop, volume in dB, keyframes), to check a change or start from the current value. | No | No |
| `timeline_export_formats` | Lists the export formats and presets of your editor. | No | No |
| `timeline_export` | Exports the timeline, a range or the In / Out to video or audio files on your disk, in one file or one per block or clip. Never overwrites a file. | No (writes files) | No |
| `timeline_update_settings` | Changes the frame size and frame rate of a sequence. | Yes | No |

### Project

| Tool | What it does | Changes your project | Uses credits |
| - | - | - | - |
| `project_read_bins` | Lists the bins and items of the project. | No | No |
| `project_search` | Searches the project by name, type or metadata. | No | No |
| `project_create_sequence` | Creates a sequence. | Yes | No |
| `project_duplicate_sequence` | Duplicates a sequence. | Yes | No |
| `project_create_bin` | Creates a bin. | Yes | No |
| `project_rename_items` | Renames project items. | Yes | No |
| `project_move_items` | Moves items to a bin. | Yes | No |

### PremiereGPT features

| Tool | What it does | Changes your project | Uses credits |
| - | - | - | - |
| `premierecopilot_connection_status` | Checks that the panel is open and connected, and in which editor. | No | No |
| `premierecopilot_job_result` | Gets the result of a tool that is still running. | No | No |
| `premierecopilot_run_transcription` | Transcribes media. Each media is transcribed once, then reused for free. | No | Yes |
| `premierecopilot_get_transcription` | Reads what is already transcribed. | No | No |
| `premierecopilot_jump_cut` | [Jump Cut](https://docs.premierecopilot.com/features/jump-cut.md): removes, mutes or marks the silences. | Yes | Yes |
| `premierecopilot_rough_cut_analyze` | [AI rough cut](https://docs.premierecopilot.com/features/claude-cut.md), step 1: finds retakes, off-camera talk, fillers, silences and the least essential sentences, and shows them in a card where you choose what to keep. Cuts nothing. | No | Yes |
| `premierecopilot_rough_cut_apply` | AI rough cut, step 2: applies the analysis with your choices from the card, after backing up the sequence. | Yes | No |
| `premierecopilot_smart_virals_analyze` | [Smart Virals](https://docs.premierecopilot.com/features/smart-virals.md), step 1: rates every sentence as a hook, development, payoff, call to action and viral potential. Ratings already computed are reused for free. | No | Yes |
| `premierecopilot_smart_virals_create` | Smart Virals, step 2: creates one vertical sequence per viral chosen by the assistant. | Yes (new sequences) | No |
| `premierecopilot_caption_styles` | Lists the Smart Captions styles and your saved presets. | No | No |
| `premierecopilot_captions_list` | Lists the captions already placed on the timeline. | No | No |
| `premierecopilot_captions` | Adds [captions](https://docs.premierecopilot.com/features/smart-captions.md) from the transcription, or edits placed captions. With `preview`, shows the style on the first caption over your picture first, for free, and keeps nothing. | Yes | Yes to add, no to edit or preview |
| `premierecopilot_auto_zoom` | [Auto Zoom](https://docs.premierecopilot.com/features/auto-zoom.md): adds zooms and punch-ins on the moments the assistant picks, optionally only on the lower tracks (a screen recording, not the facecam above it). | Yes | No |
| `premierecopilot_smart_sync` | [Smart Sync](https://docs.premierecopilot.com/features/smart-sync.md): lines up cameras and audio recorders on their sound. | Yes | No |
| `premierecopilot_podcast` | Podcast: multicam edit that switches cameras on whoever is speaking. | Yes | Yes |
| `premierecopilot_vibe_motion_brief` | [Vibe Motion](https://docs.premierecopilot.com/features/vibe-motion.md), step 1: the rules the assistant follows to write HTML animations (animated titles, lower thirds, kinetic text, infographics), with the style gallery. | No | No |
| `premierecopilot_vibe_motion_render` | Vibe Motion, step 2: exports the HTML animations written by the assistant to video and places them on a free track of the timeline. | Yes (adds clips) | Yes |
| `premierecopilot_upload_html` | Lets an assistant that runs commands (Claude Code) send long HTML as a file instead of copying it into the call. | No | No |
| `premierecopilot_index_clips` | Indexes video clips (shot type, colors, people, description). Asks first above 50 clips. | No | Yes |
| `premierecopilot_get_clip_descriptions` | Reads the descriptions of indexed clips. | No | No |
| `premierecopilot_get_music` | Picks the best music already in your project and its beats for the edit. | No | Yes |

> [!NOTE]
>
> Auto SFX is not available through the connector. Use it from the panel.

A tool that uses credits says so to the assistant, and the assistant asks before transcribing new media. Credits are the same as in the panel.

## Interactive cards

In clients that display interactive cards (such as Claude), some tools show a card next to their answer (opened full screen in ChatGPT for the richest ones):

| Card | Shown with | What you see |
| - | - | - |
| **Timeline** | `timeline_preview`, Jump Cut, AI rough cut, Auto Zoom, Smart Sync, Podcast | The tracks and clips of your sequence, also after an edit, with the zoom moments after Auto Zoom. |
| **Timeline frames** | `timeline_see_frame` | The exported frames, with their time. |
| **Smart Virals passages** | `premierecopilot_smart_virals_analyze` | The interest curve along the timeline and the best passages. |
| **Caption styles** | `premierecopilot_caption_styles` | The style gallery with short demo videos, and your presets. |
| **Transcription** | `premierecopilot_get_transcription` | The transcribed text, searchable. |
| **AI rough cut** | `premierecopilot_rough_cut_analyze` | The passages proposed for cutting: switch each one between KEEP and CUT, cut or keep single words, then apply. |
| **Vibe Motion styles** | `premierecopilot_vibe_motion_brief` | The animation styles gallery, to pick one or more. |
| **Long-running tool** | `premierecopilot_job_result` | The progress of a tool that takes several minutes. |

In clients without cards (Claude Code in the terminal, for example), the assistant gets the same result as text.

## Long tasks

Some tools take several minutes: transcription, captions, Smart Virals, the AI rough cut on a long video. Many assistants stop waiting for a tool after about 4 minutes, so when a tool runs longer than **200 seconds**, PremiereGPT answers "still running" with a job number and the tool keeps going in your editor. The assistant then calls `premierecopilot_job_result` until the result is ready.

* Nothing runs twice: don't ask the assistant to start the same task again while it is running.
* Follow the progress on the **MCP** page of the panel: every call appears there with a timer and its parameters.
* A result stays available for one hour after the end of the task.

## DaVinci Resolve

The connector works with DaVinci Resolve, with the limits of Resolve's scripting:

* **Premiere Pro only**: `timeline_edit_effects` (effects and clip speed), `timeline_transform`, `timeline_reframe`, `timeline_round_corners`, `premierecopilot_design_image`, and in `timeline_edit` moving a clip to another track or inserting a track under the others. In Resolve, the assistant gets a clear "not available in DaVinci Resolve yet" answer.
* **`timeline_edit`** in Resolve can delete, add (in an empty slot), enable, disable and unlink clips, and add markers and tracks. Splitting, trimming, moving clips and transitions are not available: the assistant builds a new edit with `timeline_create_edit` instead (without transitions).
* **Selected clips** can't be read in Resolve: give a time range or the whole timeline instead.
* **Jump Cut**: the **Mute** mode and the **Fade** transition are Premiere Pro only.
* **Podcast** always covers the whole timeline in Resolve.
* **Captions**: the editable Premiere Pro captions style (native) does not exist in Resolve. Use a gallery style.

## Safety

* **Edits happen at once.** There is no review screen outside the panel: the assistant is asked to tell you what will change and to get your go-ahead before anything destructive (cuts, deletions, moves, settings).
* **Automatic backup.** Smart Sync, Podcast, the AI rough cut and Jump Cut save a copy of the sequence in your project before changing it.
* **Not sure?** Ask the assistant to duplicate the sequence first (*"Duplicate this sequence and work on the copy"*), or duplicate it yourself.
* **Read-only tools** are marked as such for your assistant, which can use them without changing anything.

## Privacy

* **PremiereGPT keeps no content of connector calls**: no prompts, no scripts, no transcripts from these calls are stored on our servers. Error logs keep only the shape of the parameters (which fields were sent), never their values.
* Media needed by a feature (for example the audio for a transcription) is processed the same way as when you click the button in the panel.
* The assistant you connect receives what the tools return (clip names, timeline structure, transcribed text, frames). What it keeps is governed by that assistant's own privacy policy and settings.
* Sign-in uses OAuth: the assistant never sees your PremiereGPT password.

Read the full [privacy policy](https://premierecopilot.com/privacy).

## Disconnect

To remove access, open the connectors (or plugins, or apps) of your assistant, select **PremiereCopilot** and disconnect or remove it. For Claude Code, run:

```bash theme={null}
claude mcp remove premierecopilot
```

After disconnecting, the assistant can no longer call PremiereGPT. A sign-in that is not used for 90 days expires on its own. To cut every connection of your account at once, write to us at [admin@premierecopilot.com](mailto:admin@premierecopilot.com).

## Troubleshooting

> [accordion] “The PremiereGPT panel isn't connected”
>
> Open Premiere Pro or DaVinci Resolve, open the PremiereGPT panel and check it is signed in with the **same account** you used to connect the assistant. Then ask the assistant to try again. If both editors are open with the panel, close the panel in the one you don't want to edit.

> [accordion] “Credits are used up for today”
>
> Your credits for the day are spent. They reset daily; your plan and options are on the [pricing page](https://premierecopilot.com/pricing). Free tools (reading the timeline, editing, Smart Sync, Auto Zoom…) keep working.

> [accordion] A tool is still running
>
> Long tools answer "still running" after 200 seconds and keep working in your editor. Let the assistant fetch the result, and don't ask it to start the task again. Check the **MCP** page of the panel to see the call in progress. If the assistant says the job is unknown, the task finished more than an hour ago or our server restarted: ask it to read the timeline to see what was done before running the tool again.

> [accordion] “This link has expired” on the sign-in page
>
> The sign-in page is valid for 15 minutes. Start the connection again from your assistant.

> [accordion] Captions or the AI rough cut do nothing
>
> These tools use the speech already transcribed. Ask the assistant to transcribe the speech media first (it uses credits), then try again.

> [accordion] The assistant doesn't see a new tool
>
> An open conversation keeps the tool list it started with. Start a new conversation.

> [accordion] “Not available in DaVinci Resolve yet”
>
> The tool is Premiere Pro only. See [DaVinci Resolve](#davinci-resolve) for what works and the alternatives.

## Need help?

Join our [Discord](https://discord.gg/WyBNCbZhBD) or write to us at [admin@premierecopilot.com](mailto:admin@premierecopilot.com).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.