Skip to main content
Glama
README.md
# Modelflare MCP Server

Image and video generation for any MCP client — Claude Desktop, Cursor, Gemini CLI and more — with one Modelflare **Smart API Key** for every image and video model available to your account.

用一个 Modelflare **Smart API Key**,在任意 MCP 客户端里调用你账号可用的全部图像和视频模型。

mcp-name: io.github.castaly-ai/mcp-server

## Get a Smart API Key / 获取 Smart API Key

1. Sign up / 注册: https://modelflare.dev/sign-up?utm_source=mcp-registry&utm_medium=plugin&utm_campaign=integrations
2. Console → API Keys → **Create Smart API Key**, and pick a routing strategy (Balanced / Stability first / Low price first).

   控制台 API Keys →「Create Smart API Key」,选择路由策略(均衡 / 稳定优先 / 低价优先)。

## Configure / 配置

Add the server to your MCP client:

```json
{
  "mcpServers": {
    "modelflare": {
      "command": "npx",
      "args": ["-y", "@modelflare/mcp-server"],
      "env": {
        "MODELFLARE_API_KEY": "<your Smart API Key>"
      }
    }
  }
}
```

| Variable | Value |
|---|---|
| `MODELFLARE_API_KEY` | your Smart API Key (required) |
| `MODELFLARE_BASE_URL` | `global` → https://modelflare.dev/v1 (default) · `china` → https://origin.modelflare.dev/v1 · or a custom URL |
| `MODELFLARE_OUTPUT_DIR` | where images and videos are saved (default `./modelflare-output`) |
| `MODELFLARE_LOCALE` | `en` (default) or `zh` for error messages |

**Claude Desktop:** download `modelflare-mcp-<version>.mcpb` from [Releases](https://github.com/castaly-ai/modelflare-mcp/releases) and open it; Claude Desktop asks for the API key, base URL and output folder. `modelflare-mcp-<version>-smithery.mcpb` is the same bundle with each tool's `inputSchema` in the manifest, as Smithery requires. `node build-mcpb.mjs` rebuilds both.

## Tools / 工具

| Tool | What it does |
|---|---|
| `list_models` | Image and video models your key can call, each with the docs page holding its exact request fields. `kind: "text"` lists text models. |
| `generate_image` | Generates an image, or edits one with `edit: true`, and saves it. Gemini image models use `generateContent`; the others use the Images API. |
| `create_video` | Starts an asynchronous video task and returns its task ID. |
| `get_video_status` | Status and progress of a video task. |
| `download_video` | Saves a finished video; can wait up to 300 seconds first. |

Model-specific fields go in `parameters` (images) or `request` (videos) and are sent unchanged. The model list comes from your key, so new models appear without editing the key. Generated results are saved right away because result URLs expire.

Calls are billed to your Modelflare balance by actual usage; see https://modelflare.dev/usage-logs.

## Troubleshooting / 常见问题

- **401**: the key is missing or invalid. Create a Smart API Key at https://modelflare.dev/keys.
- **`model_not_found`**: a regular key only calls the models of its own group. A Smart API Key calls every model available to your account. If you already use a Smart API Key, that model is not available to your account right now.
- **403 `insufficient_user_quota`**: top up at https://modelflare.dev/billing.

## Links

Docs: https://docs.modelflare.dev/ · Pricing: https://modelflare.dev/pricing

Listed on the [M8ven MCP Trust Index](https://m8ven.ai/mcp/castaly-ai/modelflare-mcp).

## License

MIT © Havenbyte LLC

TDQS

A3.7/5.0

Scored across 5 tools

Disambiguation4/5

Each tool covers a distinct step: create_video (start task), get_video_status (poll), download_video (save), generate_image (image gen/edit), list_models (discovery). The only mild overlap is download_video optionally waiting for a task, which partially duplicates get_video_status, but descriptions clearly delineate roles.

Naming Consistency5/5

All five tools follow a clean verb_noun snake_case pattern (create_video, download_video, generate_image, get_video_status, list_models). The convention is applied uniformly with no deviations.

Tool Count5/5

Five tools is well-scoped for a media generation API, covering video task lifecycle, image generation, and model discovery. Each tool earns its place with no redundancy.

Completeness4/5

Core lifecycle is covered: create/status/download for video and generate/edit for image, plus model listing. Minor gaps exist (no cancel_task, no list_tasks, and no image status polling), but primary workflows are fully supported.

Maintenance

ActivityMaintained
ResponsivenessNo issues