A Model Context Protocol (MCP) server that enables Claude Desktop (and other MCP clients) to generate images, videos, music, and audio using Fal.ai models.

• Native Async API - Uses fal_client.run_async() for optimal performance
• Queue Support - Long-running tasks (video/music) use queue API with progress updates
• Non-blocking - All operations are truly asynchronous

• STDIO - Traditional Model Context Protocol communication
• HTTP/SSE - Web-based access via Server-Sent Events
• Dual Mode - Run both transports simultaneously

Image Generation:

• 🖼️ generate_image - Create images from text prompts (Flux, SDXL, etc.)
• 🎯 generate_image_structured - Fine-grained control over composition, lighting, subjects
• 🔄 generate_image_from_image - Transform existing images with style transfer

Image Editing:

• ✂️ remove_background - Remove backgrounds from images (transparent PNG)
• 🔍 upscale_image - Upscale images 2x or 4x while preserving quality
• ✏️ edit_image - Edit images using natural language instructions
• 🎭 inpaint_image - Edit specific regions using masks
• 📐 resize_image - Smart resize for social media (Instagram, YouTube, TikTok, etc.)
• 🏷️ compose_images - Overlay images (watermarks, logos) with precise positioning

Video Tools:

• 🎬 generate_video - Text-to-video and image-to-video generation
• 📹 generate_video_from_image - Animate images into videos
• 🔀 generate_video_from_video - Video restyling and motion transfer

Audio Tools:

• 🎵 generate_music - Create instrumental music or songs with vocals

Utility Tools:

• 🔍 list_models - Discover 600+ available models with smart filtering
• 💡 recommend_model - AI-powered model recommendations for your task
• 💰 get_pricing - Check costs before generating content
• 📊 get_usage - View spending history and usage stats
• ⬆️ upload_file - Upload local files for use with generation tools

• 600+ Models - Access all models available on Fal.ai platform
• Auto-Discovery - Models are fetched dynamically from the Fal.ai API
• Smart Caching - TTL-based cache for optimal performance
• Flexible Input - Use full model IDs or friendly aliases

• Python 3.10 or higher
• Fal.ai API key (free tier available)
• Claude Desktop (or any MCP-compatible client)

If you're using Claude Code, install directly via the plugin system:

# Add the Luminary Lane Tools marketplace
/plugin marketplace add raveenb/fal-mcp-server

# Install the fal-ai plugin
/plugin install fal-ai@luminary-lane-tools

Or install directly without adding the marketplace:

/plugin install fal-ai@raveenb/fal-mcp-server

Note: You'll need to set FAL_KEY in your environment before using the plugin.

Run directly without installation using uv:

# Run the MCP server directly
uvx --from fal-mcp-server fal-mcp

# Or with specific version
uvx --from fal-mcp-server==1.4.0 fal-mcp

Claude Desktop Configuration for uvx:

{
  "mcpServers": {
    "fal-ai": {
      "command": "uvx",
      "args": ["--from", "fal-mcp-server", "fal-mcp"],
      "env": {
        "FAL_KEY": "your-fal-api-key"
      }
    }
  }
}

Note: Install uv first: curl -LsSf https://astral.sh/uv/install.sh | sh

Official Docker image available on GitHub Container Registry.

Step 1: Start the Docker container

# Pull and run with your API key
docker run -d \
  --name fal-mcp \
  -e FAL_KEY=your-api-key \
  -p 8080:8080 \
  ghcr.io/raveenb/fal-mcp-server:latest

# Verify it's running
docker logs fal-mcp

Step 2: Configure Claude Desktop to connect

Add to your Claude Desktop config file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "fal-ai": {
      "command": "npx",
      "args": ["mcp-remote", "http://localhost:8080/sse"]
    }
  }
}

Note: This uses mcp-remote to connect to the HTTP/SSE endpoint. Alternatively, if you have curl available: "command": "curl", "args": ["-N", "http://localhost:8080/sse"]

Step 3: Restart Claude Desktop

The fal-ai tools should now be available.

Docker Environment Variables:

Using Docker Compose:

curl -O https://raw.githubusercontent.com/raveenb/fal-mcp-server/main/docker-compose.yml
echo "FAL_KEY=your-api-key" > .env
docker-compose up -d

⚠️ File Upload with Docker:

The upload_file tool requires volume mounts to access host files:

docker run -d -p 8080:8080 \
  -e FAL_KEY="${FAL_KEY}" \
  -e FAL_MCP_TRANSPORT=http \
  -v ${HOME}/Downloads:/downloads:ro \
  -v ${HOME}/Pictures:/pictures:ro \
  ghcr.io/raveenb/fal-mcp-server:latest

Then use container paths like /downloads/image.png instead of host paths.

pip install fal-mcp-server

Or with uv:

uv pip install fal-mcp-server

git clone https://github.com/raveenb/fal-mcp-server.git
cd fal-mcp-server
pip install -e .

1. Get your Fal.ai API key from fal.ai

2. Configure Claude Desktop by adding to:

   • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
   • Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "fal-ai": {
      "command": "fal-mcp",
      "env": {
        "FAL_KEY": "your-fal-api-key"
      }
    }
  }
}

Note: For Docker configuration, see Option 2: Docker above.

{
  "mcpServers": {
    "fal-ai": {
      "command": "python",
      "args": ["/path/to/fal-mcp-server/src/fal_mcp_server/server.py"],
      "env": {
        "FAL_KEY": "your-fal-api-key"
      }
    }
  }
}

3. Restart Claude Desktop

Once configured, ask Claude to:

• "Generate an image of a sunset"
• "Create a video from this image"
• "Generate 30 seconds of ambient music"
• "Convert this text to speech"
• "Transcribe this audio file"

Use the list_models tool to discover available models:

• "What image models are available?"
• "List video generation models"
• "Search for flux models"

You can use any model from the Fal.ai platform:

# Using a friendly alias (backward compatible)
"Generate an image with flux_schnell"

# Using a full model ID (new capability)
"Generate an image using fal-ai/flux-pro/v1.1-ultra"
"Create a video with fal-ai/kling-video/v1.5/pro"

Run the server with HTTP transport for web-based access:

# Using Docker (recommended)
docker run -d -e FAL_KEY=your-key -p 8080:8080 ghcr.io/raveenb/fal-mcp-server:latest

# Using pip installation
fal-mcp-http --host 0.0.0.0 --port 8000

# Or dual mode (STDIO + HTTP)
fal-mcp-dual --transport dual --port 8000

Connect from web clients via Server-Sent Events:

• SSE endpoint: http://localhost:8080/sse (Docker) or http://localhost:8000/sse (pip)
• Message endpoint: POST http://localhost:8080/messages/

See Docker Documentation and HTTP Transport Documentation for details.

This server supports 600+ models from the Fal.ai platform through dynamic discovery. Use the list_models tool to explore available models, or use any model ID directly.

These friendly aliases are always available for commonly used models:

You can also use any model directly by its full ID:

# Examples of full model IDs
"fal-ai/flux-pro/v1.1-ultra"      # Latest Flux Pro
"fal-ai/kling-video/v1.5/pro"     # Kling Video Pro
"fal-ai/hunyuan-video"            # Hunyuan Video
"fal-ai/minimax-video"            # MiniMax Video

Use list_models with category filters to discover more:

• list_models(category="image") - All image generation models
• list_models(category="video") - All video generation models
• list_models(category="audio") - All audio models
• list_models(search="flux") - Search for specific models

📖 Full documentation site: raveenb.github.io/fal-mcp-server

This project is part of the Luminary Lane Tools marketplace for Claude Code plugins.

Add the marketplace:

/plugin marketplace add raveenb/fal-mcp-server

Available plugins:

More plugins coming soon!

Error: FAL_KEY environment variable is required

Solution: Set your Fal.ai API key:

export FAL_KEY="your-api-key"

Error: Model 'xyz' not found

Solution: Use list_models to discover available models, or check the model ID spelling.

Error: File not found: /Users/username/image.png

Solution: When using Docker, mount the directory as a volume. See File Upload with Docker above.

Error: Generation timed out after 300s

Solution: Video and music generation can take several minutes. This is normal for high-quality models. Try:

• Using a faster model variant (e.g., schnell instead of pro)
• Reducing duration or resolution

Error: Rate limit exceeded

Solution: Wait a few minutes and retry. Consider upgrading your Fal.ai plan for higher limits.

Enable verbose logging for troubleshooting:

# Set debug environment variable
export FAL_MCP_DEBUG=true

# Run the server
fal-mcp

If you encounter a bug or unexpected behavior:

1. Check existing issues: GitHub Issues

2. Gather information:

   • Error message (full text)
   • Steps to reproduce
   • Model ID used
   • Environment (OS, Python version, transport mode)

3. Open a new issue with:

   **Error:** [paste error message]
   **Steps to reproduce:** [what you did]
   **Model:** [model ID if applicable]
   **Environment:** [OS, Python version, Docker/uvx/pip]
   

4. Include logs if available (with sensitive data removed)

📝 Open an Issue

Contributions are welcome! Please see CONTRIBUTING.md for guidelines.

We support local CI testing with act:

# Quick setup
make ci-local  # Run CI locally before pushing

# See detailed guide
cat docs/LOCAL_TESTING.md

MIT License - see LICENSE file for details.

• Fal.ai for providing the AI models
• Anthropic for the MCP specification