# fal-mcp

**Image generation for Claude — every model fal.ai has, inline in your conversation.**

ChatGPT has DALL-E built in. Claude doesn't — yet. This MCP server closes the gap, and then some: instead of one model, you get FLUX Schnell, FLUX Pro, Ideogram, Recraft, SDXL, and more. Privacy option via Venice.ai (prompts never logged).

```
You: generate an image of a red fox sitting in snow, golden hour
Claude: [image appears inline]
```

---

## Install

```bash
git clone https://github.com/yourname/fal-mcp
cd fal-mcp
pip install -r requirements.txt
```

Get a fal.ai key at [fal.ai](https://fal.ai) — free tier available.

---

## Add to Claude Code

```bash
claude mcp add fal-image python /full/path/to/fal-mcp/server.py --env FAL_KEY=your-key
```

Test it: start a new Claude Code session and say *"generate an image of a moonlit forest"*.

---

## Add to Claude Desktop

Edit `claude_desktop_config.json` (find it via Claude Desktop → Settings → Developer):

```json
{
  "mcpServers": {
    "fal-image": {
      "command": "python",
      "args": ["/full/path/to/fal-mcp/server.py"],
      "env": {
        "FAL_KEY": "your-fal-key-here"
      }
    }
  }
}
```

Restart Claude Desktop. A small plug icon appears when the MCP is active.

---

## Smoke test (no Claude needed)

```bash
FAL_KEY=your-key python server.py --test "a red fox in snow"
# saves test-output.jpg in current directory
```

---

## Models

| Key | Model | Best for |
|-----|-------|----------|
| `flux-schnell` | FLUX Schnell | Fast generation, everyday use (default) |
| `flux-dev` | FLUX Dev | Higher quality, worth the wait |
| `flux-pro` | FLUX Pro | Best FLUX quality |
| `flux-realism` | FLUX Realism | Photorealistic images |
| `ideogram` | Ideogram v2 | Text inside images, logos, typography |
| `recraft` | Recraft v3 | Design, vector, illustration style |
| `sdxl` | Stable Diffusion XL | Classic, reliable |
| `sana` | Sana | Fast high-resolution |

Just ask Claude to use a specific model: *"generate with recraft, a minimalist poster of..."*

---

## Privacy mode (Venice.ai)

Venice routes generation through decentralized GPUs and does not log prompts.
Add `VENICE_KEY` to your env — Claude will automatically route privacy-sensitive
requests through Venice when you ask for it.

*(Venice support: coming in v1.1)*

---

## Usage examples

Once installed, just talk to Claude naturally:

- *"Draw me a portrait of a Victorian octopus with a monocle"*
- *"Generate 4 variations of a logo concept: geometric mountain, minimal, dark background"*
- *"Make an image of Einstein riding a bicycle through Aarau, impressionist style"*
- *"What models are available?"* → Claude lists them
- *"Use flux-pro for the best quality version of..."*

---

## Cost

fal.ai bills per image. Rough guide:
- FLUX Schnell: ~$0.003/image
- FLUX Dev: ~$0.025/image  
- FLUX Pro: ~$0.05/image

Free tier available at fal.ai. No subscription required.

---

## Why not just use ChatGPT?

You're already using Claude. This gives you image generation without switching apps,
with your choice of model instead of one locked option, and with the option to keep
your prompts private.

---

## License

Do with it what thou willst. (MIT)