# fal-mcp-server

Personal MCP server for fal.ai — gives Claude Desktop, Claude Code CLI, Cursor and Windsurf full access to fal.ai's 1,000+ AI models.

## Why this exists

fal.ai's hosted MCP server requires OAuth 2.0, which Claude Desktop does not support yet. This local server solves that — your API key stays on your machine, no OAuth needed.

## Tools available

| Tool | What it does |
|---|---|
| search_models | Search fal.ai catalog by keyword or category |
| get_model_schema | Get input/output parameters for any model |
| get_pricing | Check the cost of a model before running it |
| recommend_model | Describe your task, get model recommendations |
| search_docs | Search fal.ai documentation |
| run_model | Run any model and wait for the result |
| submit_job | Submit a long-running job (video/3D) |
| check_job | Check status, fetch result, or cancel a job |
| upload_file | Upload a file to fal.ai CDN |

## Requirements

- Python 3.11+
- A fal.ai API key: https://fal.ai/dashboard/keys

## Installation

git clone https://github.com/avieirox/fal-mcp-server.git
cd fal-mcp-server
pip install -e .
echo "FAL_API_KEY=your_key_here" > .env

## Setup Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "fal-ai": {
      "command": "fal-mcp",
      "env": {
        "FAL_API_KEY": "your_fal_api_key_here"
      }
    }
  }
}

## Setup Claude Code CLI

claude mcp add --transport stdio fal-ai fal-mcp

## License

MIT