SiliconFlow MCP (aleph-siliconflow-mcp)

Aleph's official SiliconFlow media-generation MCP server: image / video / TTS. Aleph 官方的硅基流动多媒体生成 MCP 服务（图片 / 视频 / 语音合成）。

SiliconFlow ships no official MCP, so Aleph builds one. Text chat / embedding / rerank are already handled by Aleph core, so this server is media-only.

Prerequisites / 前置要求

1. Node.js 18 or newer (provides npx):

   node --version   # must be >= 18

   npx downloads and launches the server on demand — no manual install needed.
2. A SiliconFlow API key — get one at https://cloud.siliconflow.cn/account/ak.

Installation / 安装

Option A — In Aleph (recommended) / 在 Aleph 中（推荐）

This server is a built-in MCP preset in Aleph (alongside 高德地图 / 火山引擎 veImageX / MiniMax / Context7). No manual config needed:

1. Open the Aleph Panel → Settings → MCP (设置 → MCP).
2. Find 硅基流动 SiliconFlow in the preset catalog and click Install / 安装.
3. Paste your API key (and optionally a save directory for images/audio).
4. Done — the tools become available to Aleph immediately.

You can also just ask Aleph in natural language, e.g. "装一下硅基流动 MCP / install the SiliconFlow MCP", and Aleph will run the install for you (it will ask for the API key).

Option B — Claude Code / 其他支持 MCP 的客户端（CLI）

claude mcp add siliconflow \
  -e SILICONFLOW_API_KEY="your_api_key_here" \
  -e SILICONFLOW_IMAGE_DIR="/path/to/save" \
  -- npx -y [email protected]

Option C — Claude Desktop / any MCP client (JSON) / 通用 JSON 配置

Add this to your client's MCP config (Claude Desktop: claude_desktop_config.json; others: their mcpServers block):

{
  "mcpServers": {
    "siliconflow": {
      "command": "npx",
      "args": ["-y", "[email protected]"],
      "env": {
        "SILICONFLOW_API_KEY": "your_api_key_here",
        "SILICONFLOW_IMAGE_DIR": "/path/to/save"
      }
    }
  }
}

Pinned vs latest / 固定版本与最新版: [email protected] pins a reproducible release from npm (recommended); use aleph-siliconflow-mcp (no @version) to always get the latest. The first launch downloads and caches the package (a few seconds).

Configuration / 配置

| Variable | Required | Default | Description | |----------|----------|---------|-------------| | SILICONFLOW_API_KEY | Yes | — | SiliconFlow API key | | SILICONFLOW_API_BASE | No | https://api.siliconflow.cn/v1 | Override API endpoint (use .com for overseas) | | SILICONFLOW_IMAGE_DIR | No | — | Local directory to save generated images and videos | | SILICONFLOW_AUDIO_DIR | No | SILICONFLOW_IMAGE_DIR | Local directory to save generated audio |

Note / 注意: Generated media URLs expire quickly — images within ~1 hour, videos sooner. Set SILICONFLOW_IMAGE_DIR to save assets locally and avoid losing them. 生成的媒体 URL 很快过期（图片约 1 小时，视频更短），建议设置保存目录以本地留存。

Tools / 工具

| Tool | Description | |------|-------------| | generate_image | Generate an image from a text prompt using a SiliconFlow image model | | edit_image | Edit or transform an existing image (img2img) using a text prompt | | generate_video | Submit a video generation job and poll until it completes (synchronous) | | submit_video_generation | Submit a video generation job and return immediately with a request ID | | get_video_status | Poll the status of a previously submitted video generation job | | generate_speech | Convert text to speech (TTS); voice is model:voice_id | | list_models | List available models (type: image / video / audio / text; sub_type: text-to-image, text-to-video, speech-to-text, …) | | get_user_info | Retrieve account info and credit balance (total / charged / gift) |

Development / 开发

cd siliconflow
npm ci            # install dependencies
npm test          # run the vitest suite
npm run typecheck # tsc --noEmit
npm run build     # compile src/ → dist/

Note: this package was previously published on PyPI (uvx aleph-siliconflow-mcp, v0.1.0). It has been rewritten in TypeScript and now ships on npm; the PyPI release is deprecated.

License: MIT.