anime-sd-mcp

An MCP server that lets any AI client (Claude Desktop, Claude Code, Cursor, …) generate anime images through a local Stable Diffusion XL backend such as Animagine XL.

It is a thin protocol layer — it does not load the model itself. You keep running your own SD HTTP server (on your GPU); this exposes it to AI clients over the standard MCP protocol.

AI client (Claude / Cursor / ...)  --MCP-->  anime-sd-mcp  --HTTP-->  your SD backend (GPU)

Showcase

_{Generated locally and returned straight into the chat via generate_anime_image.
Left prompt: 1girl, silver hair, blue eyes, school uniform, cherry blossoms, smile}

Why

Plenty of people run a local Stable Diffusion server but can only reach it from one custom script. This wraps that server in MCP so any AI client can call it as a tool — "draw me a cat girl with silver hair" just works, inside the chat.

Requirements

• Python 3.10+
• A running local SD backend that accepts a JSON POST and returns PNG bytes (see SD backend contract below). The reference setup is an Animagine XL 4.0 server on http://127.0.0.1:7860.

Install

pip install anime-sd-mcp

Or from source:

git clone https://github.com/haotongliu58-sudo/anime-sd-mcp
cd anime-sd-mcp
pip install -e .

Configure your AI client

Claude Desktop / Claude Code

Add to your MCP config (claude_desktop_config.json, or via claude mcp add):

{
  "mcpServers": {
    "anime-sd": {
      "command": "anime-sd-mcp",
      "env": {
        "SD_SERVER_URL": "http://127.0.0.1:7860"
      }
    }
  }
}

If you installed from source without the console script, use:

{
  "mcpServers": {
    "anime-sd": {
      "command": "python",
      "args": ["-m", "anime_sd_mcp.server"],
      "env": { "SD_SERVER_URL": "http://127.0.0.1:7860" }
    }
  }
}

Restart the client. You should now see the anime-sd tools available.

Tools

| Tool | Description | | --- | --- | | generate_anime_image | Generate an anime image from Danbooru-style tags. Returns a PNG. | | check_sd_server | Check whether your local SD backend is reachable. |

generate_anime_image

| Argument | Type | Default | Notes | | --- | --- | --- | --- | | prompt | str | — | Danbooru-style tags. Required. | | negative_prompt | str | sensible default | Tags to avoid. | | width | int | 832 | Image width (px). | | height | int | 1216 | Image height (px). | | steps | int | 25 | Denoising steps. | | guidance_scale | float | 7.0 | Prompt adherence. | | add_quality_tags | bool | true | Append masterpiece-grade quality tags. |

Prompt tip: comma-separated tags work best, e.g. 1girl, silver hair, blue eyes, school uniform, cherry blossoms, smile.

Configuration (environment variables)

| Variable | Default | Description | | --- | --- | --- | | SD_SERVER_URL | http://127.0.0.1:7860 | URL of your local SD backend. | | SD_TIMEOUT | 600 | Request timeout in seconds (SDXL is slow). |

SD backend contract

anime-sd-mcp expects a backend that:

• accepts POST <SD_SERVER_URL> with a JSON body: {"prompt", "negative_prompt", "width", "height", "steps", "guidance_scale"}
• responds with raw PNG bytes (Content-Type: image/png) on success.

Any server matching this contract works. A minimal Animagine XL reference server is a few dozen lines of diffusers + http.server.

License

MIT