flux-replicate-mcp
A simple MCP server for generating images using Flux models via the Replicate API.
README
Flux Replicate MCP Server

A simple Model Context Protocol (MCP) server for generating images using Flux models via the Replicate API.
✨ Simplicity First
This server has been designed with simplicity as the primary goal:
- Minimal setup - Just add your API key and start generating images
- Zero configuration - Works out of the box with sensible defaults
- Platform-aware - Automatically organizes your images in the right place
- Essential features only - Image generation that just works, without complexity
- Easy integration - Drop into any MCP client with a single command
🚀 Quick Start
Global Installation (Recommended)
The easiest way to get started is with npx or bunx - no installation required!
# Set your Replicate API token
export REPLICATE_API_TOKEN="r8_your_token_here"
# Run with npx (Node.js)
npx flux-replicate-mcp
# OR run with bunx (Bun)
bunx flux-replicate-mcp
CLI Arguments
The server supports comprehensive CLI configuration:
# Basic usage with API key
flux-replicate-mcp --api-key r8_your_token_here
# Full configuration example
flux-replicate-mcp \
--api-key r8_your_token_here \
--model flux-1.1-pro \
--format jpg \
--quality 95 \
--working-directory ~/MyImages
# Get help
flux-replicate-mcp --help
Available CLI Arguments:
--api-key/-k/--replicate-api-key: Replicate API token (required)--model/-m: Default model (flux-2-pro,flux-2-max,flux-2-flex,flux-2-dev,flux-2-klein,flux-1.1-pro,flux-pro,flux-schnell,flux-ultra)--format/-f: Output format (jpg,png,webp)--quality/-q: Quality setting (1-100)--working-directory/-d/--dir: Custom working directory--help/-h: Show help message
📖 Complete Installation Guide →
Local Development
- Install Dependencies
bun install
- Configure Environment
cp .env.example .env
# Edit .env and add your REPLICATE_API_TOKEN
- Build and Run
bun run build
bun run start
The server will automatically create a platform-specific working directory for your generated images:
- Windows:
%USERPROFILE%\Documents\FluxImages - macOS:
~/Pictures/FluxImages - Linux:
~/Pictures/FluxImages(fallback:~/flux-images)
🔧 Configuration
All configuration is done via environment variables or CLI arguments:
| Variable | CLI Argument | Required | Default | Description |
|---|---|---|---|---|
REPLICATE_API_TOKEN |
--api-key |
✅ | - | Your Replicate API token |
FLUX_DEFAULT_MODEL |
--model |
❌ | flux-2-pro |
Default model |
FLUX_OUTPUT_FORMAT |
--format |
❌ | jpg |
Default output format |
FLUX_OUTPUT_QUALITY |
--quality |
❌ | 80 |
Default quality for lossy formats (1-100) |
FLUX_WORKING_DIRECTORY |
--working-directory |
❌ | Platform-specific | Custom working directory |
🎨 Supported Models
Flux 2 Series (Recommended)
| Model | Cost per Image | Speed | Quality | Best For |
|---|---|---|---|---|
flux-2-pro |
$0.030 | Medium | Highest | Professional work, detailed images (default) |
flux-2-max |
$0.080 | Slow | Ultra High | Premium quality, final outputs |
flux-2-flex |
$0.060 | Medium | High | Flexible, general purpose |
flux-2-dev |
$0.012 | Fast | Good | Development, experimentation |
flux-2-klein |
$0.003 | Fast | Good | Quick iterations, budget-friendly |
Flux 1 Series
| Model | Cost per Image | Speed | Quality | Best For |
|---|---|---|---|---|
flux-1.1-pro |
$0.040 | Medium | Highest | Professional work, detailed images |
flux-pro |
$0.040 | Medium | High | General purpose, balanced quality |
flux-schnell |
$0.003 | Fast | Good | Quick iterations, testing |
flux-ultra |
$0.060 | Slow | Ultra High | Premium quality, final outputs |
🛠️ Available Tools
generate_image
Generate images using Flux models with cost tracking.
Parameters:
prompt(required): Text description of the image to generateoutput_path(optional): Absolute file path for the generated image. If not provided, auto-generated filename will be used in server working directory.model(optional): Flux model to use (default:flux-2-pro)width(optional): Image width in pixels (default: 1024)height(optional): Image height in pixels (default: 768)quality(optional): Image quality for lossy formats (1-100, default: 80)
Examples:
Auto-generated filename with cost tracking:
{
"prompt": "A serene mountain landscape at sunset"
}
Response includes: file path, generation time, model used, and cost ($0.040 for flux-1.1-pro)
Custom absolute path:
{
"prompt": "Professional product photo of a smartphone",
"output_path": "/absolute/path/to/smartphone.png",
"model": "flux-pro",
"width": 1024,
"height": 1024,
"quality": 95
}
Fast iteration with flux-schnell:
{
"prompt": "Quick concept art of a robot",
"model": "flux-schnell",
"output_path": "/home/user/images/robot_concept.jpg"
}
Only $0.003 per image - perfect for rapid prototyping
Output Organization:
- Auto-generated: Files saved with descriptive names based on prompt and timestamp in server working directory
- Custom path:
output_pathmust be an absolute path for the generated image - Path validation: Relative paths are rejected to ensure compatibility across client/server environments
- Directory creation: Output directories are automatically created if they don't exist
- Cost tracking: Every generation shows the cost and model used
🎯 Design Philosophy
This server follows the principle: "Simple enough to understand in 30 minutes, powerful enough to generate great images"
What's Included
- ✅ Core image generation with Core Flux models
- ✅ Image processing and format conversion
- ✅ Platform-specific working directories
- ✅ CLI argument support with comprehensive help
- ✅ Cost tracking for budget awareness
- ✅ Basic error handling and logging
- ✅ MCP protocol compliance
🔗 MCP Integration
Claude Desktop (Recommended)
Add to your Claude Desktop MCP configuration:
{
"mcpServers": {
"flux-replicate": {
"command": "npx",
"args": ["flux-replicate-mcp"],
"env": {
"REPLICATE_API_TOKEN": "your_token_here"
}
}
}
}
Cursor Integration
Method 1: Using mcp.json
Create or edit .cursor/mcp.json in your project directory:
{
"mcpServers": {
"flux-replicate": {
"command": "env REPLICATE_API_TOKEN=YOUR_TOKEN npx",
"args": ["-y", "flux-replicate-mcp"]
}
}
}
Method 2: Manual Configuration
- Open Cursor Settings → MCP section
- Add server with command:
env REPLICATE_API_TOKEN=YOUR_TOKEN npx -y flux-replicate-mcp - Restart Cursor
Other MCP Clients
The server works with any MCP-compatible client:
- Cline: Use the same npx command
- Zed: Add to MCP configuration
- Custom clients: Use the MCP SDK
📖 Complete Integration Guide →
🚨 Error Handling
The server uses simple error codes with helpful messages:
AUTH: Authentication/API key issuesAPI: Replicate API errorsVALIDATION: Invalid input parametersPROCESSING: Image processing failures
All errors are logged as structured JSON to stderr for MCP compatibility.
💰 Cost Management
Track your spending with built-in cost reporting:
- Each generation shows the exact cost
- Model pricing clearly displayed
- Choose models based on budget vs quality needs
- Use
flux-2-kleinfor cheap iterations ($0.003) - Use
flux-2-maxfor premium results ($0.080)
📦 Installation & Usage
Global Installation
# Install globally
npm install -g flux-replicate-mcp
# Or use directly with npx
npx flux-replicate-mcp --api-key YOUR_TOKEN
# Or use with bunx
bunx flux-replicate-mcp --api-key YOUR_TOKEN
Package Information
- Package Name:
flux-replicate-mcp - Binaries:
flux-replicate-mcp,flux-replicate-mcp-server - Dependencies: 3 runtime dependencies
- Size: ~600KB unpacked
📝 Development
Build
bun run build
Development Mode
bun run dev
Publish to npm
# Build and publish
bun run build
npm publish
🤝 Contributing
- Fork the repository
- Create a feature branch
- Make your changes
- Add tests if applicable
- Submit a pull request
📞 Support
📄 License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。