Make UGC from Claude.Ask for a video. Approve the price. Get the MP4.
The ppl.studio MCP server gives your AI assistant the whole studio — workflows, creators, product photos, scripts, Veo clips. Add one URL as a connector in claude.ai, ChatGPT or Cursor and sign in — or run it locally to save every result to your disk.
https://ppl.studio/mcpMake a UGC video for https://glowlab.co/serum with Mia. Show me the preview before you render anything.
The preview is ready: 3 frames, hook “I tried five serums this year — one stayed.” Rendering costs about $1.20 on your Gemini key. Go ahead?
Go.
Done — your video is ready: final.mp4, 1080×1920 with captions burned in, plus the 3 clips and frames. It's in your ppl.studio gallery too.
How it works
- Step 01
Connect once
Paste https://ppl.studio/mcp into your app's connectors and sign in to ppl.studio — or add the local server with one line when you want files on your disk.
- Step 02
Ask in plain words
“Make a UGC video for this product page with Mia.” The assistant starts the workflow, waits for it, and shows you the frames, the lines and the price.
- Step 03
Approve, and it's yours
Say go and the video renders. You get the link to the final MP4 — and with the local server, the MP4, clips, frames and script in ~/ppl-studio.
Connect
Hosted — nothing to install
https://ppl.studio/mcp in claude.ai, Claude Desktop, ChatGPT, Claude Code, Cursor, VS Code or Windsurf. Sign in with your ppl.studio account. Works with links; 47 tools.
Local — files on your disk
One npx command in any client that starts local servers. Reads the files you name, saves every result to a folder. Needs Node 20+ and an API key from Account → API keys; 50 tools.
Name: ppl.studio
URL: https://ppl.studio/mcpIn claude.ai or Claude Desktop, add a custom connector with this URL, click Connect, sign in to ppl.studio and allow access. No key to copy — the connection gets its own key, listed in Account → API keys.
Local server settings
| Variable | Default | What it does |
|---|---|---|
| PPL_API_KEY | — (required) | Your ppl.studio API key. Without it the server starts and every tool explains how to get one. |
| PPL_OUTPUT_DIR | ~/ppl-studio | Where saved runs and downloads go. A relative folder you name lands under it; an absolute one (~/Desktop/ads) is used as is. |
| PPL_BASE_URL | https://ppl.studio | Only for testing against another ppl.studio deployment. |
50 tools over the whole studio
Grouped the way the app is. Anything without a dedicated tool is still reachable: api_catalog finds any of the API's 184 endpoints and api_request calls it.
Workflows
Run any of the 19 workflows end to end and save everything it made.
start_run · wait_for_run · approve_run · edit_run · save_run
Engine
One step on its own: animate a photo with Veo, voice a script, lip-sync, join clips.
list_bricks · start_job · wait_for_job
Pictures
UGC photos with your creator and product, new angles, Amazon and Shopify packs.
generate_image · edit_image · photo_pack · bulk_images
Words
Scripts, hooks, ad scripts, captions, translation — and a critic that ranks them.
write_scripts · write_ad · script_tools · briefs
Library
Creators (generated or yours), products, briefs, saved scripts, storyboards.
personas · products · gallery · storyboards
Files & the rest
Local server: files in, results out. Both: any of the API's 184 endpoints by name.
upload_file · download · api_catalog · api_request
Things to ask it
With the local server, paths on your machine work anywhere a file goes; with the hosted one, use links.
- Make a TikTok UGC video for https://yourstore.com/products/… with a Gen-Z skincare creator. Show me the preview first.
- Upload ~/Desktop/bottle.png as a product and make the Amazon listing photo pack. Save it to ~/Desktop/amazon.
- Write five hooks for our protein bar, pick the strongest, and turn it into a 30-second ad script.
- Animate ~/Pictures/mia.jpg saying “this changed my mornings” as an 8-second clip and save it here.
- Clone the format of this carousel that went viral, for our app. Three variants.
- What failed in my queue this week? Retry anything that can be retried.
Teach Claude how UGC gets made here
The ppl.studio skill is a short playbook Claude loads when you ask for UGC: which of the 19 workflows fits the request, what a good brief needs, how to show you the priced preview and wait for your yes, and how to use the API with curl when no MCP server is connected.
In claude.ai: Settings → Capabilities → Skills → upload the zip. Read SKILL.md first if you like.
curl -fsSL https://ppl.studio/downloads/ppl-studio-skill.zip -o /tmp/ppl-studio-skill.zip \
&& unzip -o /tmp/ppl-studio-skill.zip -d ~/.claude/skillsSafe to leave with an assistant
It asks before it spends
Video is the only expensive step. Runs stop at a priced preview, approve_run is marked as the spending tool, and the server tells the assistant to get your OK on the price first.
Your disk stays yours
The hosted server can't touch your files at all. The local one reads only the paths you give it and writes only to your output folder (~/ppl-studio by default).
Your key, your account
It acts as you through your own API key: your creators, your products, your runs. Revoke the key and it stops.
One key per connection
Signing in from claude.ai, ChatGPT or Cursor gives that connection its own key — see it and revoke it in Account → API keys. The local server is readable JavaScript you can inspect before you run it.
Questions
Hosted or local — which should I use?
Hosted if you use claude.ai, ChatGPT or Claude Desktop, or want nothing to install: paste https://ppl.studio/mcp and sign in. Local if you want the assistant to use files on your machine (product photos, your own clips) and save results into a folder — it runs with one npx command. The tools are the same.
What is the ppl.studio skill?
A Claude skill: a short playbook Claude loads when you ask for UGC. It knows which workflow fits which request, the brief fields, how to handle the priced preview, and how to use the API with curl when no MCP server is connected. Install it in Claude Code with one command, or upload the zip in claude.ai.
What is an MCP server?
The Model Context Protocol (MCP) is the open standard AI assistants use to call tools. An MCP server is a small program that gives an assistant — Claude, Cursor, VS Code's agent — a set of tools. The ppl.studio MCP server gives it 50 tools over ppl.studio: workflows, creators, photos, scripts, Veo clips and your files.
Which apps does it work with?
The hosted server (https://ppl.studio/mcp) works in claude.ai, Claude Desktop and ChatGPT as a custom connector, and in Claude Code, Cursor, VS Code and Windsurf by URL — you sign in to ppl.studio, no key to copy. The local server runs in any client that starts local servers, including the Codex CLI: npx -y https://ppl.studio/downloads/ppl-studio-mcp.tgz with your key in PPL_API_KEY.
Do I need to write code?
No. You install it once with one line or one config block, then ask in plain words: “make a UGC video for this product page”. The assistant picks the tools, shows you the preview and price, and saves the results.
What does it cost?
The MCP server is free. It uses your ppl.studio plan (10 free photos, or $1.99 a week for unlimited generation) and your own Gemini key for generation, billed by Google with no markup — about $0.045 a photo and about $0.40 for an 8-second Veo Lite clip.
Where are the videos saved?
Everything is in your ppl.studio gallery, and the assistant gives you the links. With the local server they are also saved to ~/ppl-studio (or PPL_OUTPUT_DIR, or a folder you name): final.mp4, each clip, each frame, the voice takes, script.txt and run.json.
Can it spend money without asking?
Runs stop at a priced preview and the assistant is told to ask you before approve_run. The only way around it is straight_through, which the assistant is told to use only when you explicitly say so. Photos and text are cheap and run without the stop.
Can I use my own photos and clips?
Yes. Anywhere a tool takes an image, clip or audio URL you can give a path on your machine; it is uploaded for you (images up to 5 MB, video and audio up to 512 MB).
Does it work on Windows?
Yes. The hosted server works anywhere. For the local one you need Node 20 or newer, and some clients on Windows need npx wrapped in cmd: use "command": "cmd" and put "/c", "npx" in front of the args.
One minute to your first UGC from Claude
Add the connector, sign in, ask for a video. 10 free photos to start, no card.
Free · hosted or local · claude.ai, ChatGPT, Claude Code, Cursor and more