image2-mcp

MCP Server for the company image2 text-to-image API. Allows Claude Code, Codex, and other MCP-compatible AI tools to generate images directly.

🚀 Quick Start (One Command)

git clone <repo-url>
cd image2-mcp
bash scripts/setup.sh

The script will:

1. Check that uv is installed

2. Install Python dependencies (uv sync)

3. Ask for your API key (or reuse from ~/.claude/.env)

4. Write ~/.claude/.env with your key

5. Add image2 to your ~/.claude/.mcp.json

6. Run a health check to verify everything works

Then restart Claude Code and try: "帮我生成一张图片..."

Related MCP server: Nano-Banana MCP Server

📦 Manual Setup

Prerequisites

• Python 3.10+

• uv (Python package manager)

• A company API Key for the unified AI platform

Install

uv sync

Configuration

Copy the environment template and fill in your API key:

cp .env.example ~/.claude/.env
# Edit ~/.claude/.env and set MAGENE_API_KEY

MCP Configuration

Add to ~/.claude/.mcp.json:

{
  "mcpServers": {
    "image2": {
      "command": "bash",
      "args": [
        "-c",
        "set -a; [ -f ~/.claude/.env ] && . ~/.claude/.env; [ -f .env ] && . ./.env; set +a; uv run --directory /path/to/image2-mcp python -m image2_mcp"
      ],
      "description": "文生图 — 调用公司统一 API 平台 image2 模型生成图片"
    }
  }
}

Health Check

uv run python -m image2_mcp --health-check

🔧 How It Works

The AI calls generate_image as a native tool:

AI: generate_image(prompt="a cat in a garden", size="1024x1024")
    → Server validates parameters
    → POST to company API
    → Decodes Base64 response
    → Saves PNG to disk
    → Returns image + file path + usage stats to AI

📐 Tool Parameters

Available Sizes

Custom sizes supported: WxH format, max 3840px per side, multiples of 16, aspect ratio ≤ 3:1, total pixels 655,360–8,294,400.

🧪 Development

uv sync --group dev
uv run pytest --cov=src --cov-report=term-missing