MeiGen AI Design MCP

What jau123/MeiGen-AI-Design-MCP MCP server does

jau123/MeiGen-AI-Design-MCP MCP server exposes nine tools for visual design workflows. Agents can turn short descriptions into detailed image prompts, search a curated gallery of roughly 1,446 prompts, inspect gallery inspiration, list current image and video models, and submit or monitor generations. Image generation supports MeiGen Cloud, local ComfyUI, and OpenAI-compatible endpoints. Video generation uses MeiGen models and requires a model ID.

The server also stores user preferences such as a default style, aspect ratio, model, style notes, and favorite prompts. ComfyUI users can manage workflow templates by listing them, viewing parameters, importing files, changing settings, or deleting templates.

How it works

An agent can begin with search_gallery when the user needs visual direction or has not supplied a concrete idea. get_inspiration returns the selected gallery prompt and image URLs, while enhance_prompt expands a brief concept into a more developed generation prompt. Those results can then be passed to generate_image; gallery image URLs may be used as reference images for style guidance.

list_models supplies the available model capabilities, which is especially important for video because supported tiers, resolutions, durations, and reference-video ranges are live values rather than fixed assumptions. The server validates and prices video combinations before dispatch. Long-running jobs can be checked with check_generation using the generation ID, avoiding a duplicate submission that could consume credits again.

The project also includes parallel batch generation through sub-agents and a standalone CLI, although these are separate usage modes rather than MCP tools described in the tool list.

Setup and configuration

The package can be run locally with npm using the meigen package. The README gives this MCP configuration:

{
  "mcpServers": {
    "meigen": {
      "command": "npx",
      "args": ["-y", "meigen@1.4.0"]
    }
  }
}

A hosted Streamable HTTP endpoint is also available at https://www.meigen.ai/api/mcp. Its read-only tools can be used without a token, while authenticated operations use a bearer token. For local operation, the setup wizard lets the user choose MeiGen Cloud, ComfyUI, or an OpenAI-compatible API and enter the corresponding endpoint or credentials. Local ComfyUI can run without a cloud provider, using the user's own GPU.

For the CLI mode, the README documents MEIGEN_API_TOKEN for MeiGen authentication and MEIGEN_OUTPUT_DIR to change where generated images are saved. The default output directory is ~/Pictures/meigen/.

Tools and capabilities

• enhance_prompt: expands a short idea into a detailed image prompt.
• search_gallery: finds semantically similar prompts and returns image URLs for visual browsing.
• get_inspiration: retrieves a gallery entry's full prompt and image URLs.
• list_models: reports available image and video models and capabilities.
• generate_image: submits image jobs through the configured backend.
• generate_video: submits validated MeiGen video jobs.
• check_generation: checks an existing MeiGen generation by ID.
• manage_preferences: reads or updates design preferences.
• comfyui_workflow: manages ComfyUI workflow templates.

Limitations and notes

Video generation can be slow and uses credits, so the README advises confirming before submission and avoiding parallel video jobs. If a video or image request is interrupted, use check_generation and retry polling rather than submitting the same job again. Midjourney style references use a supplied --sref code; image-based references should instead be passed through referenceImages. The remote endpoint returns media URLs, whereas local package usage can provide automatic local file saving and ComfyUI bridging.