Pictory MCP Server

The Pictory MCP server lets Claude create videos through your Pictory account. It exposes the Text-to-Video API as a set of tools that Claude calls on your behalf, so you describe the video in plain English and Claude writes the script, designs the scenes, finds the visuals and music, submits the render, and hands you the link. There is nothing to install. You add the server to Claude once, sign in with your Pictory account, and it is available in every conversation. Every action runs against your own account, so your plan, quota and AI Credits apply exactly as they do for a direct API call.

​

Two ways to connect

​

What you can ask for

Once connected, you can say things like:

• “Make a 60-second explainer about our new expense-tracking feature, professional tone, for YouTube.”
• “Create a vertical Instagram teaser for our coffee brand: warm, cozy, upbeat music.”
• “Turn this blog post into a video: https://example.com/post”
• “Show me a preview first.” Claude builds the storyboard and gives you a preview link before anything is rendered.
• “Make the title smaller and swap the clip on scene 2.” After a preview, edits are applied in place; the same preview link shows the change.
• “Which voices do you have?”, “How much quota do I have left?”

Claude settles a few things with you before it starts: what the video is about, whether it is narrated and in which voice, whether it has music, what shape it is, whether to save it as a project in your Pictory workspace, and whether you want to see a preview first or go straight to the finished video. When you design the scenes together, it also asks whether to use stock or AI-generated media, and, if you want music, whether one track plays under the whole video or a mix of tracks is chosen scene by scene. It never renders a preview into a final video without your go-ahead.

​

Who designs the scenes

This is the first question Claude asks, and the most consequential. It decides what you supply, how much control you keep, and which tools come into play. There is no default, so Claude always asks; the choice applies to the whole video, not to individual scenes.

​

Pictory designs it

Pictory’s AI Storyboard lays out every scene: the visuals, the on-screen text, the colors and the pacing. You supply the source material only, as any one of:

• A brief. A plain-language prompt for a video Pictory writes the script for. Claude asks what the video is about, what it is for (an explainer, marketing, internal communication, a tutorial or a product video), how long it should run (8–500 seconds) and how it should sound. It asks where the video will be published too, because that settles the shape.
• A script you already have. The storyboard splits it into scenes and designs each one around your words.
• A blog or article URL. Pictory reads the page and turns it into a video.
• Scene-by-scene narration. You give the narration for each scene, with an optional background per scene.

Beyond the source, you still choose the narration voice, the music, the shape, a brand kit and a subtitle style. A background you name for a scene is honored; everything else about the layout is the storyboard’s to decide, which is the point of this mode. The storyboard splits each scene’s narration at sentence ends, so an AI-generated background you name is generated once for every sentence and draws AI Credits that many times; Claude reports the estimated total with the job. Templates, smart layouts and the 4:5 ratio are not available here. Best for speed, for turning existing written content into video, and for anyone who does not want to art-direct. Searching for media is unnecessary, because the storyboard sources its own visuals.

​

Design it together

You and Claude compose each scene explicitly, then send the finished design. Per scene you control:

• Narration. It becomes both the voice-over and the burned-in subtitles. A title or call-to-action card may omit it, but then needs a minimum duration. A scene carried by an AI-generated clip may omit it too, and lasts as long as the clip (see AI-generated media).
• Background. Exactly one of a stock clip or image (from a search, or your own URL), a solid color, or an AI-generated visual. Plus an optional color tint, a slow pan on stills, and loop or mute on footage. A stock clip loops by default, so a short clip still covers the narration over it.
• Elements, up to 20 per scene, drawn in order:
  • Text overlays with control over font, size, color, case, alignment, wrap width, placement and entry or exit animation. Every text element enters with an animation unless you ask for none: a heading drifts up, other text fades in.
  • Shape overlays from a library of badges, arrows, quote marks, underlines, blobs and speech bubbles, plus plain rectangles, circles and lines that can be drawn to any width and height. A rectangle with both is a bar, a panel or a divider, which is how bar charts, stat rings and stat tiles are built. A filled shape placed before a text element gives the text its own panel.
  • Extra image or video panels, from a search or your own URL. AI-generated media always goes on the scene background, never on an element.
• Placement. An anchor preset (nine positions relative to the scene, kept clear of the edges) or an exact top and left offset in percent.
• Scene transition, minimum duration, and the option to hide subtitles where your own text elements already carry the words.
• Music, as one track for the whole video or a mix chosen scene by scene (see Music).

Claude asks whether you want to see a preview first or render straight away. A preview lets you check the design, and edits on it cost nothing; rendering straight away skips that check, so a change afterwards means a new render. The natural rhythm is draft, preview, adjust, render. A preview takes 2–5 minutes; Claude can estimate the layout before that and flag anything that overlaps. Once the preview is up, edits to wording, size, color, placement, a clip, a tint or the music are applied in place at the same URL. Only a new scene, a new element, or a mix changed beyond what can be done in place (see Music) needs a fresh preview, and when a design is resent the narration already voiced is reused: visual and layout changes never cost a new voice, and only scenes whose wording changed, or a change of voice, speed or volume, are voiced again. Best for branded, precisely laid out or template-like videos, and for data and infographic scenes. Stock media and music are free; AI-generated media draws AI Credits.

​

AI-generated media

In Design it together, Claude asks whether the scenes use stock media, which is free, or AI-generated media, which draws AI Credits. For AI-generated media it then asks, one question at a time:

1. Images, videos, or let AI decide. With let AI decide, Claude uses a video for a scene with motion, sound or dialogue and an image for a still idea, and tells you which scenes get which.
2. Sound. For videos: whether the clips make their own sound, such as ambience or speech.
3. The model. Claude lists the models that fit your answers and the video’s shape, cheapest first, with what one clip or image costs. With let AI decide it asks for a video model and then an image model. Veo and Omni Flash make no 1:1 video, so the shape is settled first.
4. Continuity and a consistent subject. With two or more AI scenes: whether the scenes should flow into each other as one continuous shot, and whether a person or product must look the same in every scene. If you share a link to an image of it, Claude uses it as a reference: on every AI scene without continuity, and on the first scene of each continuous run, whose last frame then carries the look into the scenes after it.

Scenes carried by their clip. A scene does not need narration when its AI clip carries it: a person speaking in the clip, or its own sound, such as a busy cafe or waves on a beach. Such a scene lasts as long as the clip, has no subtitles, and is not narrated, so Claude does not ask about a voice or subtitles for it. Any music plays quietly under the clip’s sound. Keep spoken lines to about 2.5 words per second of clip. Data stays on top. Generated media never shows numbers, labels or charts; those are drawn as text and shape elements over it. Claude describes only the scene, and Pictory adds where the subject should sit so the elements land on plain background. What it costs. Before the first preview, Claude shows the total AI Credits next to your balance, with cheaper options: another model, no sound on narrated clips, or stock media. It goes ahead only when you agree. Each new preview, and each render made straight from the design, generates the media again and draws the credits again. Rendering a preview you approved uses the media that preview generated, and edits on a preview cost nothing. AI clips are generated one after another, so each one adds 3–4 minutes to a preview or render.

​

Which to choose

You can ask for a preview in either mode, and in Design it together you can keep editing that preview until it is right before anything is rendered.

​

Music

If you want music, Claude asks how it should be chosen. When you design the scenes together:

• One track for the whole video. Pictory picks a track that suits the video, or Claude finds one in Pictory’s music library once the story is final and offers it to you.
• A mix chosen scene by scene. Once the scene list is agreed, Claude groups neighboring scenes that share a mood, such as the opening, the problem and the close, searches the music library separately for each group, and offers a different track for each. A track plays on across its group without restarting. If some scenes are left without a track, Claude asks whether they play one track underneath or have no music. Every track in a mix is chosen from the library; there is no automatic pick for a scene.

When Pictory designs it, one track plays under the whole video. Music sits at a low level under narration, and under an AI clip’s own sound, unless you ask for another level. On a preview, the music of a mix is one soundtrack: in place, you can make all of it quieter, or replace it with one track that then plays under every scene, including any that had no music. A different track on some scenes, or music louder than designed, means a fresh preview, which reuses the narration already voiced.

​

Tools

Fourteen tools, in four groups.

​

Creating and rendering

​

Refining a preview

​

Finding assets

​

Account and options

​

What you get back

A finished job returns the MP4, a watch link, an embed link, the narration audio, and subtitles in both SRT and VTT.

​

Quick reference

​

What you need

Two things must be in place, or the connection fails at sign-in.

​

MCP server or Claude Skill?

Both let you create videos by describing them. The MCP server works wherever you use Claude and signs in with your Pictory account. The Text-to-Video skill runs in Claude Code with your API key and carries its own design playbook for developers who want to script the process. The Pictory for Claude overview compares the two.