Playcut v2 is coming MCP · Actor v2 · & much more
34 tools · MCP + REST API · Included in every plan

The complete AI video studio,
in your AI editor.

Playcut runs a hosted Model Context Protocol (MCP) server plus a REST API on the same tool surface. Build a custom AI actor, render an actor shoot, generate a Veo clip, clone a voice, read a brand kit — 35 tools in the catalog, 34 of them callable by any external agent. One endpoint, one credit balance, one workspace.

Get your API key How to connect

No separate API fee · Included in every plan · Pay only for credits per generation

What MCP unlocks

Generate from anywhere. Ship from your editor.

Ad volume, on tap

Ask for a batch in one Claude prompt. The agent loops the tools itself — same actor, one scene per call — and reports each task as it lands.

↳ Plan concurrency: 1–10 active jobs

Brand consistency, in the prompt

Read the kit with get-brand-kit — palette, typography, logos, voice — and fold those values into the prompts your agent writes. The actor ID holds the face.

↳ Same actor ID · same brand values

Costs you can read

Every call prices in credits, and get-credit-balance reads the balance before and after a run. A 1K image is 34 credits on Nano Banana 2, 67 on Nano Banana Pro.

↳ Per-call credit cost · one shared balance

Connect

One URL. Two ways in.

The server is hosted at https://mcp.playcut.ai/mcp — nothing to install. Sign in with your browser, or send an API key from workspace settings for scripts and CI.

Claude · ChatGPT · Cursor

Custom connector · browser sign-in, no keys to copy

Settings → Connectors 
# Add a custom connector / MCP server, transport: Streamable HTTP
https://mcp.playcut.ai/mcp

# Your browser opens a Playcut sign-in — approve, and the tools appear.
# Calls run as YOUR user in your workspace.

Claude Code

CLI agent runtime · one command

claude mcp add 
claude mcp add --transport http playcut https://mcp.playcut.ai/mcp

# The first tool call triggers the browser sign-in automatically.

Any client · API key

For CI, service accounts, and hosts that support custom headers

Bearer pk_live_… 
{
  "mcpServers": {
    "playcut": {
      "type": "http",
      "url": "https://mcp.playcut.ai/mcp",
      "headers": { "Authorization": "Bearer pk_live_…" }
    }
  }
}

n8n · Zapier · custom backend

Not MCP-aware? The same tools over REST.

api.playcut.ai 
# Execute one tool — the body is the tool's input schema
curl -X POST https://api.playcut.ai/api/v1/tools/generate-image-from-text/execute \
  -H "Authorization: Bearer pk_live_…" \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Editorial portrait of …", "aspectRatio": "4:5"}'

# GET /api/v1/tools lists every tool with its JSON Schema.
Complete tool catalog

35 tools. 8 categories.

This is the shipped surface, with the same descriptions your client sees. 34 of the 35 are public — the exact list GET /api/v1/tools returns and the MCP server registers; the one marked internal backs the in-app chat agent only. On every generation tool your agent names the model explicitly or takes the documented default; nothing is chosen for it.

AI Actors

6 tools

Create an actor once, then keep the same face across stills and performance video. Six tools cover the loop: generate, list, clone a template, edit metadata, shoot images, and run a talking performance. There is no separate actor-video tool — animate an actor still with the video tools below.

generate-actor

Create a custom AI actor in the workspace library from a name plus optional descriptors (sex, age, hair). Returns the actor ID and a task ID; an initial photo shoot is chained automatically to populate the profile image.

list-actors

List the workspace's actors, the curated template actors, or both. Use the returned ID with actor-shoot and actor-act.

clone-actor-template

Duplicate a curated template actor into your workspace library so you own an editable copy, then use it like any actor you generated yourself.

update-actor

Edit an actor's metadata — rename, update description. It triggers no regeneration; to change traits, create a new actor with generate-actor.

actor-shoot

Generate images of an actor in a scene and outfit — product holds, lookbook stills, brand compositions. Takes scene/outfit asset IDs or text descriptions; 1:1 by default, up to 4K, with 2×2 / 3×3 / 4×4 grids rendered inside one image.

actor-act

Multi-step performance: Playcut generates the scene, then animates the actor delivering the prompt. The video engine speaks the quoted dialogue itself — AI voice, no voice ID accepted. Four tiers via engine: fastest, fast, plus, pro.

Images

2 tools

Two image tools, both on Google's Gemini image models. Your agent passes an explicit model id or takes the default — nothing is selected for it. grok-imagine-image is offered in the Playcut studio but is not selectable through these MCP tools.

generate-image-from-text

Generate an image from a text prompt. Models: gemini-3-pro-image-preview (Nano Banana Pro, default — highest fidelity, best in-image text rendering) or gemini-3.1-flash-image-preview (Nano Banana 2 — roughly half the credit cost, and the only model that accepts the 512 budget resolution).

generate-image-from-reference

Generate an image from a prompt plus 1–14 reference asset IDs — put a product in a scene, inherit a style, match an existing asset. Same two Gemini models, Nano Banana Pro by default.

Video

6 tools

Six video tools. Each takes an optional model id and defaults to veo-3.1-generate-preview; per-model duration, resolution, and aspect-ratio windows are enforced server-side before any credits are held. Eleven aspect ratios and 480p–4K are in the schema, narrowed per model.

generate-video-from-text

Text-to-video. Models: veo-3.1-generate-preview (default), grok-imagine-video, seedance-2.0, happyhorse-1.1, gemini-omni-flash-preview, wan-2.7. Duration windows differ per model — Veo 4/6/8s, Seedance 4–15s, Omni Flash fixed at 10.

generate-video-from-image

Animate one still. Provide a motion prompt plus a single image asset ID; defaults to veo-3.1-generate-preview at 720p, 16:9.

generate-video-from-reference

Prompt plus 1–9 reference images anchoring style or subject, with optional per-reference weights. Per-model caps: Veo 3, Grok 7, Seedance 9, Wan 5, HappyHorse 9, Omni Flash 6.

generate-video-extension

Continue an existing video by another 8 seconds at 720p — both values are locked by the Veo extension constraint. Pass the source video asset ID and a prompt for what happens next.

generate-video-interpolation

Generate the motion between two still frames. veo-3.1-generate-preview is locked to 8s; seedance-2.0 runs 4–15s and up to 4K but rejects frames containing human faces.

generate-video-edit

Transform an existing video with a prompt — style change, subject swap, scene transform. wan-2.7 (default) takes a 2–10s source; happyhorse-1.1 takes 3–60s (first 15s used) plus up to 5 reference images. Billed on source and output seconds.

Voice & Speech

4 tools

Text-to-speech, voice cloning, and voice design. The language field accepts English, Chinese, Spanish, French, German, Italian, Japanese, Korean, Portuguese, Russian, or Auto. Designed and cloned voices join the workspace library and work in every later TTS call.

generate-tts

Synthesize up to 5,000 characters with any voice ID from list-voices. Stock voices accept a free-text prompt steering tone, pace, and emotion; cloned voices ignore it to preserve acoustic identity.

clone-voice-from-audio

Clone a voice from 10–60 seconds of clean single-speaker speech. Returns the voice ID plus a task ID; the voice is usable in generate-tts once cloning completes.

design-voice

Synthesize a brand-new voice from a free-text description — no audio sample needed. Returns the voice ID plus a task ID.

list-voices

List voices available to the workspace: curated presets, designed, cloned, or all three. Filter by language and status.

Assets & Files

6 tools

Every generation lands in the workspace as an asset. List, fetch, upload, move, share, and delete — the same operations the app performs. Signed URLs from get-asset expire in about an hour; share-asset returns a stable public URL instead.

get-asset

Full detail for one asset — type, MIME, signed URL (valid ~1 hour), thumbnail, file info, provenance, and parent folder, brand kit, or session.

list-assets

List assets with filtering by type, folder, brand kit, session, or actor. Returns lightweight summaries; call get-asset for signed URLs.

upload-asset

Upload an image, video, or audio file from base64 bytes (~20MB cap) or a remote URL. Lands in the folder you name or the caller's default landing folder.

move-asset

Re-parent an asset into a folder or brand kit. Session-attached assets can't move — pass copyMode: 'copy' to leave the original alone and drop a copy in the target.

share-asset

Create a public share link. The asset is copied to the public bucket and served from an unguessable URL that doesn't expire.

delete-asset

Soft-delete an asset. It disappears from listings and is removed permanently by the daily retention cron — recoverable until then, typically 24–48h.

Folders & Sessions

4 tools

Folders decide who can see the work; sessions thread related generations together. Create either from an agent, then pass the folder ID on upload or the session ID as workerId on a later generation call.

create-folder

Create a folder for organizing sessions and assets. PRIVATE is creator-only; TEAM is workspace-shared.

list-folders

List folders. Each user has exactly one default PRIVATE folder; the workspace has at least one default TEAM folder.

create-session

Create a session — STUDIO, BOARD, or ACTOR_SESSION. Returns the session ID to pass as workerId on later generation calls.

list-sessions

List the workspace's sessions with metadata — used to pick a workerId for the next call.

Workspace & Brand Kits

5 tools

Multi-brand brand kits, plan context, and credit visibility. Your agent can read which brand it is working on and how much budget is left before it queues anything.

get-current-workspace

Active workspace info — plan, member count, active-subscription flag, owner. Useful for plan-aware behavior before a large batch.

get-brand-kit

Full brand kit detail — colors, typography, logos, and brand voice (tone, do-say, don't-say, brand story). Feed it into prompts to stay on-brand.

list-brand-kits

List the workspace's brand kits. Agencies run one per client brand.

get-credit-balance

Current credit balance — total, subscription, purchased, and usage this period. Check it before queueing a batch.

list-workspaces Internal

List every workspace the signed-in user belongs to, with role, plan, and which one is active. Internal: it gives the in-app chat agent its context and is not exposed to external MCP or REST clients, which are scoped to one workspace by the X-Workspace-Id header.

Tasks

2 tools

Every generation returns a task ID. There is no blocking wait tool on the public surface — poll get-task-status until the task reaches COMPLETED, FAILED, or CANCELLED, then fetch the asset.

get-task-status

Current status, progress, and output for a task. Output is populated on COMPLETED with signed-URL-enriched data.

list-tasks

List tasks filtered by status, type, or session. Lightweight summaries; call get-task-status for full output.

Real workflows

What teams are automating today.

Slack → a batch of UGC ad variants

01/04

A marketer types a brief in Slack. A Claude agent calls list-actors to find the actor, get-brand-kit for the palette and voice, then fans out actor-shoot calls and animates the winners with generate-video-from-image. It polls get-task-status and posts each finished file back with its credit cost.

// Claude prompt
"Shoot @mira holding the new serum in 10 scenes,
9:16, then animate the three best as 8s clips"

Shopify CSV → e-com catalog

02/04

A merchandiser drops a Shopify product CSV into a Cursor workspace. An agent uploads each product photo with upload-asset, calls generate-image-from-reference per SKU, and files the results into a per-campaign folder created with create-folder — ready for Shopify upload.

// Cursor agent script
products.forEach(p =>
  playcut.call("generate-image-from-text", {
    prompt: p.scene,
    model: "gemini-3.1-flash-image-preview",
    aspectRatio: "1:1",
    imageSize: "1K",
  })
)

Abandoned-cart flow → personalized recovery ads

03/04

An n8n workflow fires on cart-abandonment events. It hits the REST tool endpoint to shoot the brand's actor holding the specific abandoned product, animates that still with generate-video-from-image, then ships the MP4 to Klaviyo for the recovery email.

// n8n HTTP node
POST https://api.playcut.ai/api/v1/tools/actor-shoot/execute
Authorization: Bearer pk_live_…
{
  "actorId": "665f…",
  "prompt": "Holding {{cart.product.name}} in window light",
  "aspectRatio": "9:16"
}

Agentic creative pipeline

04/04

A long-running Claude agent runs on a cron. Every Monday it audits last week's ad performance, picks the winners, re-shoots fresh variants of the top creative through Playcut, and logs the credit spend it read from get-credit-balance before and after the run.

// Claude agent (excerpt)
const winners = await meta.topAds({ days: 7 });
for (const w of winners.slice(0, 3)) {
  await playcut.call("actor-shoot", {
    actorId: w.actorId,
    prompt: w.brief,
    outputFormat: "2x2",
  });
}
Authentication

Browser sign-in. Or a bearer key.

People sign in through the browser, so tool calls run as their own user. Scripts send a workspace API key created in settings — shown once, revocable at any time, and never echoed back in a tool response.

  • OAuth 2.1 + PKCE, discovered from the server
  • Or bearer pk_live_ keys · TLS-only
  • Per-workspace scoping · multi-tenant safe
  • Concurrent jobs by plan: 1 / 3 / 6 / 10
Pricing

No separate API fee.

MCP and REST access is bundled into every Playcut plan. You pay only the credits each generation costs — the same rate as running it in the app, drawn from the same workspace balance. Those per-generation rates are listed model by model on the model comparison.

  • 01 Hobby $9/mo · 500 credits · 3 custom actors
  • 02 Pro $29/mo · 2,000 credits · 10 actors
  • 03 Studio $79/mo · 6,000 credits · 25 actors
  • ✓ Agency $149/seat/mo · 10,000 credits · unlimited actors
  • 05 All plans · API + MCP included
Questions

MCP & API FAQ

What is the Playcut MCP server? +

Playcut runs a hosted MCP server that exposes the studio as Model Context Protocol tools. Any MCP client — Claude, Claude Code, Cursor, Windsurf, Zed, ChatGPT, or a custom agent — can call generate-image-from-text, generate-video-from-text, actor-shoot, clone-voice-from-audio, get-brand-kit, and the rest. 35 tools in the catalog — 34 callable by any external client, one internal chat-agent helper. One endpoint: https://mcp.playcut.ai/mcp.

How do I connect the Playcut MCP server? +

There is nothing to install — the server is hosted. In Claude, open Settings → Connectors → Add custom connector and paste https://mcp.playcut.ai/mcp. Claude Code takes `claude mcp add --transport http playcut https://mcp.playcut.ai/mcp`, and the first tool call opens a browser sign-in. For scripts and CI, send a workspace API key as a bearer header instead.

Is there a REST API too? +

Yes, and it is the same tool surface. GET https://api.playcut.ai/api/v1/tools returns every public tool with its input and output JSON Schema; POST /api/v1/tools/{name}/execute runs one. n8n, Zapier, Make, and custom backends use the same pk_live_ keys the MCP server accepts.

Which AI clients work with Playcut MCP? +

Any client that speaks Model Context Protocol over Streamable HTTP — Claude, Claude Code, Cursor, Windsurf, Zed, ChatGPT connectors, and custom MCP hosts. Clients that support custom headers can skip the browser sign-in and pass an API key. Anything that isn't MCP-aware uses the REST endpoints.

Can I generate AI actors programmatically? +

Yes. generate-actor creates a custom actor from a name plus optional descriptors, chains an initial photo shoot, and returns an actor ID that stays valid for later calls. actor-shoot renders stills of that actor in any scene or outfit, and actor-act produces a talking performance in which the video model speaks your quoted dialogue. For plain motion, animate an actor still with generate-video-from-image.

What does pricing look like for API + MCP usage? +

There is no separate API fee — every plan (Hobby $9, Pro $29, Studio $79, Agency $149/seat) includes MCP and REST access, and you pay only the credits a generation costs. A 1K image is 34 credits on Nano Banana 2 or 67 on Nano Banana Pro. Video bills per second: 160 credits/second for Veo 3.1 at 720p, 10 for Seedance 2.0 at 480p.

How is auth handled? Is it secure? +

Two ways. A browser sign-in (OAuth 2.1 with PKCE, discovered from the server's protected-resource metadata) ties tool calls to your own user, or a workspace-scoped pk_live_ API key is sent as a bearer token. Keys are created in workspace settings, shown once, revocable at any time, and all traffic is TLS.

Are there rate limits? +

Two limits apply. Concurrent generation jobs come from your plan — Hobby 1, Pro 3, Studio 6, Agency 10 — and a submission past the cap returns 429 rather than queueing. API-key traffic also runs against a per-workspace 60-second sliding window shared by every key in the workspace, with X-RateLimit headers on each response and Retry-After on a 429.

Build the AI video workflow
your competitors can't.

Paste one URL into your client, approve the sign-in, and start automating. Every Playcut tool — actors, images, video, voice, brand kits, files — at your agent's fingertips.

Get your API key Browse the 34 tools