Updated September 2026 | 60-word TL;DR: You can generate on-brand marketing images without leaving Claude by connecting the Avocado AI MCP server. This step-by-step guide covers setup, your first generation, choosing the right model, batch workflows for ecommerce campaigns, and cost breakdowns. No extra tabs, no copy-pasting prompts between tools.
Jump to:
- What you need before you start
- Step 1: Connect Avocado to Claude
- Step 2: Generate your first marketing image
- Step 3: Pick the right model for the job
- Step 4: Build batch workflows for product campaigns
- How much does it cost?
- FAQ
- Start with Avocado AI
Why generate images inside Claude?
Most ecommerce teams open one tab for ChatGPT, another for their image tool, a third for asset management, and a fourth for scheduling. The Avocado AI MCP server collapses that. Once connected, Claude can generate images, edit them, and save them to your workspace without you switching context.
This guide walks through the full setup and your first real marketing image generation in Claude using the Avocado MCP server.
What you need before you start
- A Claude account (web or desktop). Claude Pro, Team, or Enterprise all work.
- An Avocado AI paid plan. The MCP server is included on every plan, starting at Intro (EUR 19.99 per month, 100 credits). Generations use the same credits as the web app.
- An API key from avocadoai.co/api-keys if you are connecting from Claude Code, Codex, Cursor, or another MCP client. Claude web connects via OAuth (no key needed).
Step 1: Connect Avocado to Claude
The setup takes under a minute.
From Claude web or desktop
- Open Settings, then Connectors.
- Click Add custom connector.
- Paste the endpoint URL:
https://mcp.avocadoai.co/api/mcp - Sign in with your Avocado account when prompted. Claude uses OAuth, so no API key is required.
From Claude Code
Run this in your terminal:
claude mcp add avocado-ai --transport http https://mcp.avocadoai.co/api/mcp \
-H "Authorization: Bearer YOUR_API_KEY"
Replace YOUR_API_KEY with the key from your API keys page.
Verify the connection
Ask Claude: "List the tools you have from Avocado." Claude will enumerate the available tools, including generate_image, edit_image, generate_video, generate_music, and others. If you see those, you are connected.
Step 2: Generate your first marketing image
Once connected, ask Claude in plain language. Here are real prompts that work well for ecommerce brands:
Product hero shot:
"Generate a 4:5 product hero image for our vitamin C serum. The bottle should be on a marble surface with soft morning light from the left. Clean, minimal background."
Social media ad:
"Create a 1:1 Instagram ad for our new matcha powder. Show the packaging on a wooden cutting board with scattered matcha powder and a ceramic bowl. Warm, natural light."
Lifestyle shot:
"Make a lifestyle image of someone holding our sunscreen bottle at the beach. Golden hour light. 9:16 vertical format for Stories."
What happens next
- Claude calls the
generate_imagetool on the Avocado MCP server. - Avocado picks the default model (Nano Banana 2) unless you specify one.
- The server returns a job ID. Claude can poll
check_jobto track progress. - When the image is ready (typically 10 to 30 seconds), it appears in your Avocado workspace gallery.
- Claude shows you the result in the conversation.
Every image you generate this way lands in your Avocado AI workspace with full commercial rights, no watermarks. You can edit, upscale, or drop it into a Flow later from any surface.
Step 3: Pick the right model for the job
The Avocado MCP server exposes multiple image models. Each has different strengths:
| Model | Best for | Credits per image | Max resolution |
|---|---|---|---|
| Nano Banana 2 | Fast iterations, social posts, concept testing | 1 credit | 2K |
| GPT Image 2.5 Flare | Sharp text in images, product shots with labels, up to 4K | 1 to 4 credits | 4K |
| Recraft v4.1 | Design-focused work, vectors, clean graphics | 1 credit | 2K |
| SeedDream 5.0 Pro | Photorealistic lifestyle and editorial shots | 2 credits | 2K |
| Soul 2 | Artistic, stylized visuals | 2 credits | 2K |
How to specify a model in Claude
Just name it in your prompt:
"Use GPT Image 2.5 to generate a product photo of our new lip balm packaging. 4K resolution. Studio lighting, white background."
Claude passes the model name to the generate_image tool automatically.
When to use which
- Testing concepts quickly: Nano Banana 2. One credit per image means you can iterate ten times for the price of a single high-res generation.
- Product listings with text: GPT Image 2.5 Flare. It renders labels, ingredient lists, and price tags more accurately than other models.
- Ad creatives at scale: Nano Banana 2 for volume, then GPT Image 2.5 for the final polished version.
- Social media content: Recraft v4.1 for clean, design-forward graphics.
Step 4: Build batch workflows for product campaigns
The real power of Claude plus MCP shows up when you need multiple images for a campaign. Instead of generating one at a time in a web app, you can describe the full set in one conversation.
Example: A product launch campaign
"I'm launching a new lavender hand cream. Generate these four images:
1. A clean product shot on white, 1:1 for the website hero
2. A lifestyle shot of someone applying the cream at a vanity table, 4:5 for Instagram feed
3. A flat lay with the cream, dried lavender sprigs, and a linen towel, 1:1 for Pinterest
4. A close-up texture shot showing the cream's consistency, 16:9 for the email header"
Claude will call generate_image four times, potentially in sequence. Each result lands in your workspace. You can then ask Claude to edit any of them:
"Take image 1 and swap the background to a soft lilac gradient."
The edit_image tool modifies the existing image and saves the new version alongside the original.
Using Brand DNA for consistency
If you have Brand DNA set up in Avocado (products, colors, visual rules), Claude can read it through the MCP server. This means your generations automatically follow your brand guidelines without you repeating them in every prompt.
"Generate three Instagram story backgrounds for our brand. Use our Brand DNA colors and style."
The agent pulls your saved brand context and applies it to the generation.
How much does it cost?
The MCP server uses the same credit system as the Avocado web app. No extra fees for using Claude as your interface.
| Plan | Monthly price | Credits per month | Approximate images (Nano Banana 2) |
|---|---|---|---|
| Intro | EUR 19.99 | 100 | 100 images |
| Starter | EUR 39 | 300 | 300 images |
| Growth | EUR 99 | 800 | 800 images |
| Pro | EUR 249 | 2,000 | 2,000 images |
At the Starter tier, each Nano Banana 2 image costs roughly EUR 0.13. A GPT Image 2.5 high-quality generation (4 credits) costs about EUR 0.52. Credits roll over for up to one year, and you can buy top-up packs anytime.
The agent sees the credit cost before it generates, so there are no surprises.
FAQ
Does Claude need an Avocado account to generate images?
Yes. Claude web connects via OAuth (you sign in with your Avocado account). Claude Code and other MCP clients use an API key from your Avocado dashboard. Either way, you need a paid Avocado plan.
Can I use this workflow with other AI agents?
Yes. The Avocado MCP server works with any client that supports Streamable HTTP: Claude, OpenAI Codex, Cursor, Windsurf, Hermes, and custom-built agents. Each uses the same endpoint and tools.
Where do the generated images go?
Every image is saved to your Avocado workspace gallery. You can access it from the web app, edit it in Studio, drop it into a Flows pipeline, or schedule it for publishing from any surface.
How is this different from using a standalone image generator?
Standalone generators (like using DALL-E directly or Midjourney) produce images that live in one chat. With Avocado MCP, the output is saved to a shared workspace, knows your Brand DNA, and can go straight into a Flow, a scheduled post, or a campaign. Your team can pick it up later from any surface.
Can I generate video ads the same way?
Yes. The generate_video tool exposes models like Seedance 2.5, Kling 3.0, and MiniMax H3. Ask Claude for a product video from a photo or text prompt. Video generations take one to four minutes and the agent can poll check_job until the clip is ready.
Do I need to know which model to use?
No. If you do not specify a model, Avocado defaults to Nano Banana 2 for images (fast, 1 credit). You can always name a specific model in your prompt if you want a different result, like GPT Image 2.5 for sharp text rendering or SeedDream 5.0 Pro for photorealistic shots.
How to pick in under 30 seconds
- Need fast product mockups? Use Nano Banana 2, 1 credit each.
- Need sharp text on packaging images? Use GPT Image 2.5 Flare.
- Need clean design-forward graphics? Use Recraft v4.1.
- Need photorealistic lifestyle shots? Use SeedDream 5.0 Pro.
- Working at scale? Set up Brand DNA once, then batch-generate through Claude.
- On a budget? Start with Intro at EUR 19.99/mo, 100 credits.
If you want one creative system that works inside Claude and saves everything to a shared workspace, start with Avocado AI. Check out our pricing for details. The Avocado MCP server is included on every plan and connects in under a minute.
Wanderson Jackson is the founder of Avocado AI, a creative workspace for ecommerce brands that generates on-brand images, videos, music, and ad creative at scale.