How to Give Your AI Agent Image Generation (MCP Guide, 2026)

What Makes an Agent Image-Capable?

An MCP-compatible agent becomes image-capable by connecting to an MCP server that exports image tools. MediaEngine provides those tools through both local (`npx`) and remote (https://mcp.mcpmediaengine.com/mcp) endpoints.

The key difference: **local MCP runs on your machine; remote MCP is HTTP-accessible from anywhere**.

Setup: Claude Desktop (Local)

Step 1: Get Your API Key

Sign up at mcpmediaengine.com and create an API key. You'll receive a live key like `me_live_xyz...` and 25 free credits.

Step 2: Configure Claude Desktop

Open `~/.claude/claude.json` (macOS/Linux) or `%APPDATA%/Claude/claude.json` (Windows) and add:

{
  "mcpServers": {
    "mediaengine": {
      "command": "npx",
      "args": ["-y", "mcp-media-engine"],
      "env": {
        "MEDIAENGINE_API_KEY": "me_live_your_key_here"
      }
    }
  }
}

Step 3: Restart Claude

Quit and reopen Claude Desktop. The image tools are now available.

Step 4: Generate an Image

Simply ask Claude: "Generate an image of a sunset over mountains, 16:9 aspect ratio, vibrant colors"

Claude will use the tool and display the result inline.

Setup: ChatGPT (Remote Endpoint)

For ChatGPT or cloud-based agents, use the hosted remote endpoint. No local installation needed.

Configure Your Agent

{
  "mcpServers": {
    "mediaengine": {
      "url": "https://mcp.mcpmediaengine.com/mcp",
      "env": {
        "MEDIAENGINE_API_KEY": "me_live_your_key_here"
      }
    }
  }
}

The remote endpoint works identically to local — same tools, same syntax, no code changes.

Setup: Custom Agents

For Node.js agents or custom deployments:

const client = new MediaEngineClient({ apiKey: process.env.MEDIAENGINE_API_KEY });

// Call image generation directly const result = await client.generateImage({ prompt: 'A watercolor painting of a forest', model: 'nano-banana', aspectRatio: '16:9' });

console.log(result.imageUrl); ```

Available Models

• **nano-banana**: Speed-optimized, 1 credit per image (~7¢)
• **gpt-image-2**: Higher quality, 4 credits per image (~28¢)
• **seedance**: Video-capable, 2-3 credits per frame

Aspect Ratios

1:1, 16:9, 9:16, 21:9, 9:21, 3:2, 2:3, 5:4, 4:5, 4:3, 3:4

Common Pitfalls

**"Tool not found"** — Restart Claude after editing `claude.json`.

**"Invalid API key"** — Verify the key starts with `me_live_` and matches exactly.

**"Network timeout"** — If using remote MCP, check your internet connection; remote should respond in <1s.

**Slow generation** — First generation takes ~7s (warming up). Subsequent calls are faster.

Next Steps

• Explore [async job patterns](/blog/async-media-generation-for-ai-agent-pipelines) for high-volume workflows
• Learn [cost optimization strategies](/blog/how-much-does-ai-image-generation-cost-for-an-automated-content-pipeline)
• Try [img2img style transforms](/blog/ai-image-styles-compared-watercolor-cartoon-claymation)