ProductAI in Claude — Accurate on-brand AI Product Photos and more

ProductAI transforms your product shots into realistic, high-converting AI product photos. No design skills needed. Trusted by 20,000+ brands and creators.

Now available inside Claude (Claude.ai, Claude Desktop, or Claude Code). Generate, edit, and preview product photos by chatting — no copying API keys, no leaving the conversation. Finished images appear inline so you can compare variants side by side.

hero placeholder — three sneaker variants rendered inline in a Claude chat

What you can do

"Generate a product photo of this sneaker on a clean white marble surface — give me 3 variants" → three completed images appear directly in the chat.
"Re-shoot this candle with warm evening light on a wooden table" → one variant, finished.
"Use kontext-pro to add sunglasses to this product" → swaps the model under the hood, returns the result.
"Check the status of job 245922" → reports back the live status, attaches the image when ready.
"How good are the photos on mystore.com/products/candle?" → a score out of 100 and a list of what to fix.
"Import these three products and shoot each on marble" → products land in your catalogue, then get re-shot.
"Make a template from this Pinterest pin" → a reusable scene you can put any product into.
"Did the label change in that last photo?" → the fidelity check compares it with your original.

You stay in Claude the whole time.

One-time setup (under a minute)

Fastest: click Add to Claude — Claude opens with the connector pre-filled. Confirm, sign in, done.

Or add it manually:

Open Settings → Connectors → Add custom connector in Claude.
Server URL:
```
https://mcp.productai.photo/mcp
```
Click Connect. A window opens.
Sign in with your existing ProductAI account (create.productai.photo).
That's it — Claude lists the ProductAI tools and the connector shows Connected.

You never touch an API key. The connector mints (or reuses) one on your account automatically the first time you call a tool.

setup placeholder — connectors screen showing ProductAI connected

Tools the connector exposes

Create

Tool	What it does
`generate_image`	Start a generation job (1–4 variants in parallel).
`generate_and_wait`	Start, wait, and return the finished image(s) in one step.
`generate_video`	Start a product video (Seedance 2.0 or Gemini Omni).
`get_job`	Look up a job by ID; returns the image inline when complete.
`wait_for_job`	Poll a running job until it finishes.
`upload_asset`	Upload a photo from the chat and get a hosted URL to generate from.

Workflows

Tool	What it does
`start_onboarding`	Guided first run: learns what you sell, then walks you to your first photo step by step.
`test_pdp`	Score the photos on a product page or whole store (out of 100) and list what to fix. Free.
`import_products`	Import products from product page links (up to 10) into your catalogue, photos included.
`list_products`	List the products already in your catalogue.
`create_template_from_pinterest`	Turn a Pinterest pin into a reusable scene template saved to your account.
`list_templates`	Browse scene templates by category, or your own saved ones.
`check_fidelity`	Compare a generated photo with your original and flag any change to the product (text, colour, shape, missing parts, size). Free.

You don't normally call these by name — just describe what you want and Claude picks the right one.

Example prompts

Quick re-shoot

Here's a sneaker: https://example.com/product.jpg — place it on a clean white marble surface, soft natural daylight from the left.

Compare variants

Generate 3 variants of this candle on a wooden table, warm evening lighting, slight bokeh.

Edit a product

Use kontext-pro on this image — add aviator sunglasses to the model, keep everything else identical.

Check a product page

How good are the photos on https://mystore.com/products/amber-candle? What should I fix?

Import and re-shoot

Import https://mystore.com/products/amber-candle and https://mystore.com/products/oud-candle, then put each on a dark walnut table.

Template from Pinterest

Make a template from https://pinterest.com/pin/123456789/ and use it on my amber candle.

Check the product stayed exact

Check job 270035 — did the label or colour change?

New here?

Help me get started with ProductAI.

Track a long job

What's the status of ProductAI job 270035?

Models available

Reference files (upload_asset): Attach an image in chat (or give a URL) and it's uploaded to ProductAI, returning a hosted URL you can pass as image_url to any of the tools below. Max 10MB; PNG, JPG, or WebP.

Image (generate_image / generate_and_wait): nanobanana2.1 (default), nanobananapro (best quality), gpt-2.6, gpt-2.6-precise, gpt-2.5, gpt-2.5-precise, gpt-2, gpt-1.5, gpt-low, gpt-medium, gpt-high, kontext-pro, kontext-max, nanobanana, nanobanana2, seedream, seedream-5-lite, seedream-5-pro.

gpt-2.6 — OpenAI GPT Image 2.6, the best GPT model: better images and faster (~11s) than gpt-2. 10 credits.
gpt-2.6-precise — the precise-edit variant of GPT Image 2.6: changes only what you ask for and keeps the product, composition and background as they are. Best for small targeted edits to an existing photo; slower (~25–30s). 10 credits.
gpt-2.5, gpt-2.5-precise — the previous version. They keep working for existing workflows; use the 2.6 names for new work. 10 credits.
nanobanana2.1 — Google Nano Banana 2.1, the successor to nanobanana2, and the default: most of nanobananapro at 5 credits instead of 8.

Multi-image input (passing several reference photos in one call, up to 5) works with gpt-2.6, gpt-2.6-precise, gpt-2.5, gpt-2.5-precise, gpt-2, gpt-1.5, seedream, nanobananapro, nanobanana, nanobanana2, and nanobanana2.1; seedream-5-lite and seedream-5-pro take up to 10.

Video (generate_video): Seedance 2.0. Give a prompt and, optionally, one or more reference images (up to 9) to animate a product — omit the image for text-to-video. Options: resolution (480p/720p/1080p/4k), aspect_ratio, duration (4–15s or auto), generate_audio. It returns a short video (a video URL, not a still) and takes a few minutes, so start it and check back with wait_for_job or get_job. Billed per second of output scaled by resolution (e.g. a 5s 720p clip ≈ 35 credits).

Credits, billing, and limits

Each generation deducts credits from your ProductAI balance — same as the dashboard or direct API.
Use count: 3 and you'll be charged for three generations (one per variant).
The page check (test_pdp), fidelity check, product import, templates and onboarding don't use credits. The page check is limited to 10 per hour per account.
Manage your subscription and credits at https://create.productai.photo.

Privacy & security

The connector authenticates via your existing ProductAI account through Auth0. No password is ever stored.
Image URLs you paste are uploaded once to ProductAI's processing pipeline (the same one the dashboard uses) and then handed to the model.
Generated images live in the same S3 bucket as your dashboard outputs and follow your account's retention policy.
The MCP server itself is open source: https://github.com/ShapeStudio/productai-mcp-server.

Troubleshooting

"Authorization failed" when connecting. Sign in to https://create.productai.photo first to make sure your account is active, then retry.

Connector connects but a generation never starts. Your account may be out of credits. Check create.productai.photo/dashboard.

"Could not provision a ProductAI API key for this account." Rare race condition; retry the same prompt and it'll succeed.

Want to use ProductAI from a local script instead of Claude? The original stdio version is at https://github.com/ShapeStudio/productai-mcp — same tools, no auth flow needed.

Support

Email [email protected], or open an issue at https://github.com/ShapeStudio/productai-mcp-server/issues.