# Fixing Claude’s Image Problem: Because I Can Fix Her

> Claude can't generate images — until you wire it to Gemini over MCP. A step-by-step guide to turning Claude into a multimodal agent.

Claude is a 10, but it can’t generate images.
Still, I had a feeling: *I can fix her.*

Turns out… you actually can.

Not by waiting for an official feature drop, but by wiring Claude into something that already does it well.

In this guide, we’ll plug **Gemini (Google’s multimodal model)** into Claude using MCP (Model Context Protocol), effectively turning Claude into a **multi-model agent** that can generate images on demand.

## What’s Really Happening Here

You’re splitting responsibilities across models:

- **Claude** → reasoning, instructions, orchestration
- **Gemini** → image generation
- **MCP** → the bridge that lets them talk

## Step 1: Get Your Gemini API Key

Head over to:
👉 [https://aistudio.google.com/api-keys](https://aistudio.google.com/api-keys)

- Create a new API key
- Make sure **billing is enabled** (this matters for image generation)

---

## Step 2: Plug Gemini into Claude (MCP Setup)

Now the fun part, connecting the two.

### Open Claude Desktop:

- Go to **Settings → Developer → Edit Config**
- Open `claude_desktop_config.json`

### Add this:

```json
"mcpServers": {
  "gemini": {
    "command": "npx",
    "args": ["@houtini/gemini-mcp"],
    "env": {
      "GEMINI_API_KEY": "YOUR_API_KEY_HERE"
    }
  }
}
```

> Replace "YOUR_API_KEY_HERE" with your actual key.

That’s it. You’ve just given Claude access to Gemini.

---

## Step 3: When Things Break (Because They Might)

Let’s be real, this setup isn’t always plug-and-play.

Here’s what usually goes wrong:

- Node.js version is too old → use **v18+**
- npx isn’t working properly
- API key has no billing attached
- MCP server fails silently

### What worked for me

I hit compatibility issues locally, so I Grabbed the MCP server logs and dropped them into Claude Code, and... as expected it fixed everything in minutes.

> If you’re stuck, check the logs first. Always.

---

## Step 4: Test It

Now try this:

> “Hey Claude, use the Gemini MCP to generate an image of what the world might look like in five years.”

### 💡 Pro tip

Claude won’t always decide to use tools on its own, you need to **nudge it clearly**.

---

## Final Thoughts

Claude didn’t need to change.

You just gave it access to the right tool.

And suddenly, a limitation turned into a capability.

*Yeah… you fixed her.* <d>



</d>



---

## 📚 Resources

- [https://github.com/houtini-ai/gemini-mcp](https://github.com/houtini-ai/gemini-mcp)
- [https://aistudio.google.com](https://aistudio.google.com)