Updated October 2026 | TL;DR: Claude can generate, edit, and batch-produce product images inside a single conversation when you connect it to an image-generation MCP server. This guide walks through the full workflow using Avocado AI as the MCP backend, from first setup to marketplace-ready product shots.
Jump to:
- Why MCP changes product photography
- What you need before you start
- Step 1: Connect Avocado MCP to Claude
- Step 2: Choose the right image model
- Step 3: Generate your first product image
- Step 4: Edit existing product photos
- Step 5: Batch product images at scale
- MCP vs Studio: when to use each
- Pricing and credit costs
- FAQ
Why MCP changes product photography for ecommerce brands
Traditional product photography costs $200 to $5,000 per session. AI image generators cut that to $0.10 to $2.00 per image, but most tools require you to leave your workflow, open a web app, type a prompt, download the result, and repeat for every SKU.
MCP (Model Context Protocol) removes that friction. When you connect an image-generation MCP server to Claude, you describe what you want in plain language, and Claude calls the generation tools directly. No app-switching. No manual downloads. Claude handles the prompt engineering, model selection, and result retrieval inside the conversation.
For ecommerce brands running 20 to 100 SKUs with 3 to 8 images each, that workflow difference adds up fast. A task that takes an afternoon in a web app takes 15 minutes in Claude with MCP.
The Avocado AI MCP server is one option for this. It exposes image generation, image editing, and a library of AI models through a single MCP endpoint that works with Claude, Codex, Cursor, and other MCP-compatible clients. Every plan includes MCP access, and the CLI (@avocado-ai-co/cli) provides the same tools from a terminal.
What you need before you start
- A Claude account (Claude Desktop, Claude Code, or Claude Web)
- An Avocado AI account with an active plan (see pricing, plans start at EUR 19.99/month)
- Product photos if you want to edit existing images (any background, any lighting)
That is it. No design skills, no photography background, no API configuration files. The MCP server handles model routing, generation, and delivery.
Step 1: Connect Avocado MCP to Claude
There are two ways to connect, depending on which Claude surface you use.
Option A: Claude Code (recommended for teams and automation)
Install the CLI globally:
npm install -g @avocado-ai-co/cli
Then authenticate and install the MCP server for Claude Code:
avocado auth login
avocado mcp install --client claude-code
This writes the MCP configuration to Claude Code's settings. Claude can now call all 68 Avocado tools, including image generation, editing, and model listing.
Option B: Claude Desktop or Claude Web
Add the Avocado MCP server URL directly in your MCP client settings:
https://www.avocadoai.co/mcp
Sign in when prompted. Every image generation call requires authentication, so Claude will prompt you if the session has expired.
Once connected, verify the setup by asking Claude: "List the available image models." Claude will call the models_list tool and return the current catalog.
Step 2: Choose the right image model
Not every AI image model handles product photography the same way. The Avocado MCP server gives you access to multiple models, and Claude can help you pick the right one based on what you need.
| Task | Best model | Why |
|---|---|---|
| Clean white-background catalog shots | GPT Image 2.5 Flare (default) | Consistent product rendering, clean edges, marketplace-ready |
| Lifestyle scenes with a product | Nano Banana 2 | Strong at placing products in realistic settings |
| Background removal or scene swap on existing photos | GPT Image 2.5 Flare Edit | Preserves the product while changing everything else |
| High-detail close-ups for premium brands | Recraft v4.1 | Sharp detail, good for jewelry, cosmetics, and textured surfaces |
| Quick variations for A/B testing | Nano Banana 2 | Fast,1 credit per image at 2K resolution |
Credit costs per image (approximate):
| Model | Resolution | Credits per image | Cost on Starter (EUR 39/mo, 300 credits) |
|---|---|---|---|
| Nano Banana 2 | 2K | 1 | ~EUR 0.13 |
| GPT Image 2.5 Flare | Low quality | 1 | ~EUR 0.13 |
| Recraft v4.1 | Standard | 1 | ~EUR 0.13 |
| GPT Image 2.5 Flare | High quality | 4 | ~EUR 0.52 |
On the Starter plan at EUR 39/month with 300 credits, a1-credit model works out to about EUR 0.13 per image. That is 80 to 95 percent cheaper than a traditional studio session.
Step 3: Generate your first product image
With the MCP server connected, generating a product image is a conversation. Here is a real workflow:
You: "Generate a product photo of a matte black water bottle on a marble countertop. Soft window light from the left, shallow depth of field, clean composition."
Claude calls the generate_image MCP tool with your description, selects the default model (GPT Image 2.5 Flare), and returns the result directly in the conversation.
Tips for better product images through Claude:
- Name the material. "Matte black aluminum" produces better results than "black bottle." The model uses material cues for texture and light behavior.
- Specify the background. "Clean white studio background" for marketplace listings. "Styled kitchen counter with marble veining" for lifestyle shots.
- Describe the lighting. "Soft diffused window light from the upper left" gives more control than "good lighting." Claude translates your description into a precise prompt.
- Request the aspect ratio. "16:9 landscape for a website banner" or "1:1 square for Instagram." The model respects these when specified.
- Iterate in the same conversation. If the first result is close but not right, say "make the background lighter" or "try a45-degree angle." Claude adjusts the prompt and regenerates.
Step 4: Edit existing product photos
If you already have product photos, Claude can edit them through MCP instead of generating from scratch. This is useful when you want to keep your actual product (with its exact label, shape, and details) but change the setting.
You: "Take this product photo and place the bottle on a wooden table in a cozy autumn setting. Warm tones, candlelight feel."
Claude calls the image edit tool, preserving the product while swapping the background. The edit models available through the Avocado MCP include:
- GPT Image 2.5 Flare Edit (default) - best all-around for background swaps
- Nano Banana 2 Edit - good for scene changes while keeping product fidelity
- Flux Kontext Max - strong at maintaining product details in complex scenes
Common edit workflows for ecommerce:
- Background removal: "Remove the background and place on pure white (#FFFFFF) for Amazon listing compliance."
- Seasonal variants: "Create three versions: spring (flowers, pastel background), summer (beach, bright light), and winter (cozy interior, warm lighting)."
- Scale and angle: "Show this product from a 45-degree elevated angle with a person's hand reaching for it."
Each edit costs credits based on the model used. Most edit models are 1 credit per image.
Step 5: Batch product images at scale
This is where MCP really outperforms web-app workflows. Instead of generating one image at a time, you can describe an entire batch to Claude.
You: "I have 12 skincare products. For each one, I need: one white-background catalog shot, one lifestyle scene on a bathroom vanity, and one close-up of the texture. Here are the products: [list]. Generate all36 images."
Claude breaks this into a structured workflow:
- Lists the products and confirms the image types
- Calls
generate_imagefor each product in sequence - Saves each result with a clear filename (e.g.,
serum-catalog.png,serum-lifestyle.png) - Reports the total credit cost when finished
On the Growth plan (EUR 99/month, 800 credits), generating36 images at1 credit each costs36 credits, leaving 764 credits for the rest of the month.
Pro tip: Claude can also generate images in parallel by issuing multiple MCP calls in quick succession. For large batches (50+ SKUs), the Growth or Pro plan is recommended because they support more parallel generations per user (8 on Growth).
MCP vs Studio: when to use each
Avocado AI is not a dedicated product photography tool. It is a creative workspace that covers images, video, UGC, audio, and publishing in one credit balance. The MCP server gives you programmatic access to the same image models that the Studio workbench uses. Here is when each surface makes more sense:
| Surface | Best for | Why |
|---|---|---|
| Claude + MCP | Batch generation, iterative conversations, automation, developer workflows | Describe in natural language, iterate in conversation, script for scale |
| Studio (web app) | Visual editing with presets, upscale, draw-to-edit, exploring models | Visual interface with real-time preview, preset library, manual refinement |
| Flows | Repeatable multi-step pipelines (generate, edit, upscale, export) | Build once, run for every product or brand |
| Super Agent | Autonomous content creation with publishing | Plans, creates, schedules, and publishes after your approval |
For most ecommerce product photography, Claude + MCP covers the generate-and-iterate loop well. If you need manual touch-up (painting out a logo, precise crop, or multi-layer compositing), switch to Studio.
Pricing and credit costs
Every Avocado plan includes MCP access. There is no separate MCP subscription.
| Plan | Monthly | Credits/month | MCP access | Best for |
|---|---|---|---|---|
| Intro | EUR 19.99 | 100 | Yes | Testing and low-volume catalogs |
| Starter | EUR 39 | 300 | Yes | Small catalogs (20-50 SKUs) |
| Growth | EUR 99 | 800 | Yes + Super Agent | Medium catalogs, team workflows |
| Pro | EUR 249 | 2,000 | Yes + Super Agent | High-volume production, agencies |
Yearly billing saves 20 percent. Credits roll over for up to one year. Top-ups are available anytime.
Representative costs for a20-SKU product catalog (3 images each =60 images):
| Model | Credits per image | Total for 60 images | Cost on Starter |
|---|---|---|---|
| Nano Banana 2 (2K) | 1 | 60 credits | ~EUR 7.80 |
| GPT Image 2.5 Flare (low) | 1 | 60 credits | ~EUR 7.80 |
| Recraft v4.1 | 1 | 60 credits | ~EUR 7.80 |
Compare that to a traditional product photography session: $200 to $5,000 for the same60 images.
See Avocado AI pricing for the latest tier details and credit costs.
FAQ
Can Claude generate product images directly without MCP?
No. Claude is a text-based AI. It cannot generate images on its own. MCP (Model Context Protocol) bridges the gap: Claude describes what it wants, the MCP server calls an image model, and the result comes back into the conversation. Without an MCP server connected, Claude can only write prompts for you to paste into a separate tool.
Do I need API keys to use the Avocado MCP server?
You need an Avocado account with an active plan. The MCP server authenticates through your Avocado login or an avk_ API key. There is no separate developer API or API key management beyond what the MCP server and CLI use.
Can I use Claude MCP for Amazon-compliant product images?
Yes. Ask Claude for "a product photo on a pure white background, square format, with the product filling 85 percent of the frame." The image models produce clean white-background shots that meet Amazon's main image requirements. You may still need to verify the exact RGB white value (255, 255, 255) after generation, which Claude can help check.
What if the product in the generated image does not match my real product?
AI image generation creates new images from your description. For exact product fidelity (your specific label, packaging, and details), use the image edit workflow instead: upload your real product photo and ask Claude to change the background or setting while keeping the product intact. The edit models preserve the original product appearance.
How does Claude MCP compare to dedicated product photography tools?
Dedicated tools like Pebblely, Photoroom, or Nightjar focus specifically on product photography with marketplace templates and batch processing. The Avocado MCP server is broader: it covers images, video, UGC, audio, and publishing in one workspace. For pure product photography at high volume, a dedicated tool may have more specialized templates. For brands that also need video ads, UGC content, and social publishing, the consolidated approach means one credit balance and one library instead of three or four subscriptions.
Can I use the MCP server with other AI clients besides Claude?
Yes. The Avocado MCP server works with any MCP-compatible client, including OpenAI Codex, Cursor, Hermes Agent, and OpenClaw. The setup process is similar: add the MCP server URL, authenticate, and start generating.
How to pick the right approach in under30 seconds
- New to AI product photography? Start with Avocado Studio (the web app) to explore models and presets, then graduate to Claude + MCP for batch work.
- Running20 to50 SKUs? Claude + MCP on the Starter plan handles it in a single session.
- Agency managing multiple brands? The Growth or Pro plan with Super Agent can plan, create, and schedule product content across brands.
- Need exact product fidelity? Always use the edit workflow (upload your real photo, change the background) rather than generation from scratch.
- Budget-conscious? Nano Banana 2 at1 credit per image (~EUR 0.13 on Starter) is the most cost-effective option for clean product shots.
- Want automation? Use the CLI (
@avocado-ai-co/cli) or build a script that calls the MCP server directly for fully automated product image pipelines.
If you want one creative system for product photography, start with Avocado AI. Check out our pricing for details. The MCP server, Studio, Flows, and Super Agent all share one credit balance, so you can generate images in Claude, refine them in the web app, and publish them through Super Agent without switching platforms.
About the author: Wanderson Jackson is the founder of Avocado AI, a creative workspace for ecommerce brands that combines AI image generation, video, UGC, audio, and publishing in one platform.