AI

Connect your AI

Let Claude, ChatGPT or Cursor make explainer videos and demos with Dolly.

Three ways in

There are three ways for an AI to work with Dolly. The first two are Connect your AI: your AI uses Dolly itself. The third is for teams: your AI uses the videos in your workspace.

• On your Mac: Claude Desktop, Claude Code, Cursor.One click each in Connect your AI…. Your AI records, edits and exports with Dolly over a local connection. Set it up
• On the web: claude.ai and ChatGPT.Your own connector URL reaches Dolly on your Mac while Allow web AIs is on. Dolly needs to be open. Set it up
• For your team: a workspace AI key.Claude Code, claude.ai, ChatGPT or Cursor find, read and download your team's videos, from anywhere, with no Mac. Set it up

What it does

With Dolly Pro or Lifetime, your AI assistant can use Dolly through MCP, the Model Context Protocol. Ask it for a demo and it records, cuts, zooms and exports. Ask it for explainer videos of your product and it learns the app, writes a plan, and Dolly records a narrated, captioned series with an index page. Everything happens on your Mac: apps on your Mac talk to Dolly over a local connection, and Claude or ChatGPT on the web through your own connector URL. For AI agents, the one-page overview is llms.txt.

On your Mac

Open Dolly from the menu bar (or File, or Dolly Home) and choose Connect your AI…. The AI apps on your Mac are listed first, each with one Connect button, and Connect all does them at once. The page says Connected as soon as the app reaches Dolly.

• Claude Desktop: click Connect; Claude shows the Dolly extension; click Install. The extension uses the copy of Dolly in Applications, so Dolly updates need no reinstall.
• Cursor: click Connect; Cursor asks to install the dolly MCP server; click Install.
• Claude Code: click Connect. Dolly adds itself to Claude Code for every project and installs the explainer skill. No Terminal needed.

If you move Dolly out of Applications, the page says Needs update; click Reconnect.

Claude on the web

claude.ai (and Claude on your phone, with the same account) reaches Dolly through your own connector URL. In Connect your AI…, turn on Allow web AIs (it needs your Dolly account), then:

1. Click Copy URL & open Claude. Claude opens at Customize › Connectors.
2. Click + › Add custom connector. Name it Dolly, paste the URL and click Add.
3. Click Connect. On your Mac, Dolly asks Allow Claude to use Dolly? Click Allow.
4. In a chat, click + › Connectors and turn on Dolly.

On Claude Free you can add one custom connector. On Team and Enterprise an owner adds it under Organization settings › Connectors.

Dolly MCP reaches your Mac through our relay, so it's part of Dolly Cloud, in Pro.

ChatGPT

ChatGPT on chatgpt.com, and the ChatGPT app, use the same connector URL. Turn on Allow web AIs, then:

1. Click Copy URL & open ChatGPT. ChatGPT opens at Plugins.
2. Click + › Add custom MCP server. Name it Dolly, paste the URL as the Server URL and keep OAuth.
3. Tick I understand and want to continue, click Create as a plugin, then Install. Dolly asks Allow ChatGPT to use Dolly? Click Allow.
4. In a chat, type @ and pick Dolly.

Your workspace admin may need to allow custom MCP servers.

How web AIs reach your Mac

Dolly keeps one outgoing connection to link.getdolly.app while Allow web AIs is on; nothing on your Mac listens to the internet. Each new app needs your Allow in Dolly, even with the URL. From the web, Dolly writes and reads files only in ~/Movies/Dolly, and putting a video on your website stays a job for an AI on your Mac. New URL disconnects every web app at once; Disconnect removes one. Requests are rate-limited.

While your AI works

A small AI at work panel appears at the top left: which AI it is, what it's doing (Recording your screen…, Exporting 1080p… 42%), the step of the plan with a live picture, and the last steps with checkmarks. Stop ends it. When a video is ready the panel says Ready with Open, then fades. The panel is never in a recording and never takes focus. Explainers of a website are recorded in Dolly's browser behind your other windows, so you can keep working; videos of a Mac app use your real mouse, and the panel says so first. Videos an AI made carry a small AI mark in Dolly Home.

By hand

Claude Desktop or Cursor: add this to ~/Library/Application Support/Claude/claude_desktop_config.json or ~/.cursor/mcp.json, then restart the app:

{"mcpServers": {"dolly": {
  "command": "/Applications/Dolly.app/Contents/MacOS/Dolly",
  "args": ["/Applications/Dolly.app/Contents/Resources/mcp/dolly-mcp.mjs"],
  "env": {"ELECTRON_RUN_AS_NODE": "1"}
}}}

Claude Code:

claude mcp add dolly --scope user -e ELECTRON_RUN_AS_NODE=1 -- \
  /Applications/Dolly.app/Contents/MacOS/Dolly \
  /Applications/Dolly.app/Contents/Resources/mcp/dolly-mcp.mjs

And the skill:

mkdir -p ~/.claude/skills/dolly-explainers && cp /Applications/Dolly.app/Contents/Resources/skills/dolly-explainers/SKILL.md ~/.claude/skills/dolly-explainers/

Explainer videos

Ask: “Make explainer videos of my app with Dolly. It runs at http://localhost:3000.” (In Claude Code you can also run /mcp__dolly__make_explainers.) Your AI then:

• reads Dolly's guide (dolly_explainer_guide) and learns your product from the code, the docs and the running app;
• writes the plan, dolly-explainers.md, in plain Markdown: one task per video, with what the narrator says;
• tries every step in Dolly's own browser window without recording;
• makes the series (dolly_make_series): every video with the same look, in ~/Movies/Dolly/Explainers/<series>, plus posters and an index.html;
• reviews contact sheets of each video and retakes what's off.

## Connect your first app
Audience: new admins
Goal: connect Slack in under a minute
Voice: auto
Music: Still Water
Steps:
1. say: Here's how to connect your first app.
2. click: "Apps"
3. click: "Connect Slack"
4. wait for: "Connected"
5. say: Done. Slack is now searchable.

Steps: say, click, type … into …, select … in …, press, scroll, focus, wait for, go, hover. Web apps are recorded in Dolly's own browser window, so your mouse stays free. In Mac apps, Dolly uses the real mouse and keyboard, which needs Accessibility permission.

Narration

Voice: auto uses the best voice on your Mac, spoken by Apple's own voices, free and offline. Captions follow the voice, and music steps back while it talks. For the most natural sound, download a free Premium voice once: System Settings → Accessibility → Spoken Content → System voice → Manage Voices… (English: Ava, Zoe, Evan, Nathan). Voice: off gives captions only.

Updating a series

Say “Update my Dolly explainer videos.” Your AI runs dolly_check_explainers, which replays each video's steps and compares the screens, fixes the plan where a button was renamed, and re-records only what changed, under the same file names.

Just ask your AI

Say it the way you would to a colleague. Each of these works today, and each is tested end to end:

• “Put my latest Dolly video as the first video on our website's home page.”dolly_list_projects → dolly_publish_web into the site's public/videos → your AI pastes the hero snippet at the top of the home page and shows you the page.
• “Replace the demo video in our README with a GIF of my last recording.”dolly_export with format gif and a path in your repo → the README's image points at it.
• “Turn my last recording into a clickable demo and put it on our pricing page.”dolly_edit op demo → dolly_export with format demo into the site's public folder → an iframe on the pricing page.
• “Add my latest video to the changelog post for v2.3, with chapters.”dolly_describe → dolly_edit op describe (title, chapters) → dolly_publish_web → the player snippet, with captions and chapter tracks, in the post.
• “Cut the ums and long pauses from my last video, add a title and chapters, and give me a link.”dolly_transcript clean (a dry run first) → dolly_describe → dolly_edit op describe → dolly_get_link.
• “Make a vertical version of my last video with captions, ready for Reels and TikTok.”dolly_edit op captions → dolly_export with also_vertical: a 9:16 copy that follows the action.
• “Use the Midnight look for every video I record from now on.”dolly_looks apply with id midnight and default: true.
• “Give me a link to my last recording with the password “launch” that expires in 7 days.”dolly_get_link with settings: password launch, expires_days: 7.
• “Which videos got the most views this week, and what are people asking in the comments?”dolly_link list, then get per link: viewers in the last seven days from Insights, and the comments.
• “Re-record our onboarding explainer videos. The app changed.”dolly_check_explainers → fix the plan → dolly_make_series with only_changed: same file names.
• “Write a help article from my last recording with the steps I took.”dolly_get_project (clicks and typing) and dolly_transcript → dolly_export with format demo (a still and caption per step) → the article.

Not yet through your AI: translating captions (do it in the Captions tab) and title cards on a recording (in the editor).

On your own website

Ask “Put my latest Dolly video as the first video on our website's home page.” in Claude Code or Cursor, in your site's repo. Your AI calls dolly_publish_web with your site's public folder (for example public/videos), and Dolly writes:

• The video: an H.264 MP4 that starts playing right away (fast start), 1080p at 30 fps at a web bitrate.
• A silent copy for a video that plays by itself: without a sound track, browsers autoplay it on every device.
• A poster as JPG and WebP, so there's a picture before the video loads.
• Captions and chapters as WebVTT, when the recording has a transcript and chapters.

It returns the width, height and length and ready-to-paste code for plain HTML, React and Next.js, and Vue: one version that autoplays muted on a loop, one with sound and controls. Both keep the right shape while loading, so the page doesn't jump. If the recording already has a Dolly link, you also get the embed code. Your AI puts the code in the page and shows you the result.

Dolly only writes into the folder your AI names, which must already exist, and never replaces a file unless asked. Nothing is uploaded; committing and deploying stay yours. In Claude Code you can also run /mcp__dolly__publish_to_website.

Everything else in Dolly

Your AI can use every part of Dolly, through the same editor you use, so every change is undoable with ⌘Z:

• Transcript (dolly_transcript): read what was said, cut or restore words by their text, hide words from the captions, and Remove ums and long pauses, with a dry run first.
• Looks (dolly_looks): apply a built-in look, one of your presets or a look link, save it as a preset, make it the look every new recording starts with, or share a project's look.
• Teleprompter (dolly_script): set the script, pace, size and theme, and show or hide it.
• Annotations and drawings (dolly_edit): add, change or remove text, arrows, boxes and numbered steps; change, hide or bring back the pen drawings you made while recording.
• Interactive demos: dolly_export with format demo (a folder) or demo_link, and dolly_edit op demo for the title, captions and steps. See Interactive demos.
• iPhone and iPad: dolly_list_sources lists one on a cable and recording takes source: "phone".
• Dolly links (dolly_link): list your links, read insights and comments, change the password, expiry, downloads, comments, button, title and description, reply to a comment as the creator, pick the poster frame, or replace the video under the same link. dolly_get_link takes the same settings.

What stays yours: drawing with the pen during a take, and deleting links or comments. Uploading happens only when you ask for a link.

Tools

dolly_explainer_guide, dolly_plan, dolly_browser, dolly_look, dolly_make_video, dolly_make_series, dolly_check_explainers, dolly_voices, dolly_review_video, dolly_record_demo, dolly_start_recording, dolly_stop_recording, dolly_pause_recording, dolly_list_sources, dolly_auto_edit, dolly_add_zoom, dolly_edit, dolly_transcript, dolly_looks, dolly_script, dolly_export, dolly_list_projects, dolly_get_project, dolly_describe, dolly_publish_web, dolly_get_link, dolly_link, dolly_open_project and dolly_status. Prompts: make_explainers, update_explainers, polish_recording, publish_to_website.

Free and Pro

On Free, the read-only tools (status, sources, your projects with their transcript, look, demo steps, links and script, and title suggestions), the explainer guide and plan, the voice list and Auto edit answer; everything that records, changes, exports or uploads through your AI needs Dolly Pro. Nothing is uploaded unless you ask your AI for a Dolly link.