MeiGen AI Design MCP

Image and video generation, visual edit workflows and prompt-gallery search on MeiGen

Official · jau123 TypeScript Remote (hosted) · Streamable HTTPLocal · stdio AI and MLMediaDesign

75 popularity score ★ 1.8k ↓ 513/wk

Connect your client

Remote (hosted)

Add in one click

Opens the app, which asks before it adds the server.

Or add it by command or config

ChatGPTSettings → Apps → Advanced → Developer mode, then Create app
https://www.meigen.ai/api/mcp
Claude (desktop and web)Settings → Connectors → Add custom connector
https://www.meigen.ai/api/mcp
Cursor~/.cursor/mcp.json
{
  "mcpServers": {
    "meigen-ai-design": {
      "url": "https://www.meigen.ai/api/mcp"
    }
  }
}
Codexterminal
codex mcp add meigen --url https://www.meigen.ai/api/mcp
openclawterminal
openclaw mcp add meigen --url https://www.meigen.ai/api/mcp --transport streamable-http --auth oauth --timeout 240
Local (npm)

Add in one click

Opens the app, which asks before it adds the server.

Or add it by command or config

Claude CodeTerminal
claude mcp add meigen-ai-design -- npx -y meigen@3.4.0
Claude Desktopclaude_desktop_config.json
{
  "mcpServers": {
    "meigen-ai-design": {
      "command": "npx",
      "args": [
        "-y",
        "meigen@3.4.0"
      ]
    }
  }
}
Cursor~/.cursor/mcp.json
{
  "mcpServers": {
    "meigen-ai-design": {
      "command": "npx",
      "args": [
        "-y",
        "meigen@3.4.0"
      ]
    }
  }
}
VS Code.vscode/mcp.json
{
  "servers": {
    "meigen-ai-design": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "meigen@3.4.0"
      ]
    }
  }
}
Codex~/.codex/config.toml
[mcp_servers.meigen-ai-design]
command = "npx"
args = ["-y", "meigen@3.4.0"]
Gemini CLITerminal
gemini mcp add --scope user meigen-ai-design npx -- -y meigen@3.4.0

Settings 9

MEIGEN_API_TOKEN secret Optional: Existing MeiGen API key. Gallery search, prompt rewriting and preferences work without it. Can also be set in ~/.config/meigen/config.json. For MeiGen Cloud generation, video and visual workflows.
MEIGEN_OUTPUT_DIR Optional: Where generated images are saved. On Linux, $XDG_PICTURES_DIR/meigen is used when that is set. Default ~/Pictures/meigen.
MEIGEN_VIDEO_OUTPUT_DIR Optional: Where generated videos are saved. Default ~/Movies/meigen. For generate_video.
MEIGEN_REQUEST_STORE_DIR Optional: Private directory for job receipts used to recover interrupted jobs. Default ~/.meigen/requests.
UPLOAD_GATEWAY_URL Optional: Upload gateway for reference files. Can point to your own presign endpoint. Default https://gen.meigen.ai.
MEIGEN_BASE_URL Optional: MeiGen API origin. Default https://www.meigen.ai.
Set together: openai-provider
OPENAI_API_KEY secret Optional: Key for an OpenAI-compatible provider used instead of MeiGen for image generation. For generate_image.
OPENAI_BASE_URL Optional: Base URL of the OpenAI-compatible provider. Default https://api.openai.com.
OPENAI_MODEL Optional: Image model ID on that provider. Default gpt-image-2.

What it does

This server connects an assistant to MeiGen, a paid service for AI image and video generation. You can search MeiGen's public gallery of example images along with the prompts behind them, check which models are on offer and what they cost, then submit image or video jobs and poll them until they finish. Five fixed visual workflows sit on top of that: background removal, upscaling, product detail shots, marketing posters, and placing a product on a new background. Local reference images and short audio or video clips can be uploaded so the video models can use them.

There are two ways to use it. The makers now point new users at the hosted endpoint at https://www.meigen.ai/api/mcp, which signs in with a MeiGen account over OAuth. Only that connection can see a user's favorites, boards, history and saved characters. The npm package `meigen` runs locally over stdio and stays for people who already hold a MeiGen API key, or who want plain image generation through their own OpenAI-compatible API. Prompt rewriting and a stored preferences file work locally without any key.

The local package registers 18 tools (src/server.ts). The newest is render_video_draft. Some video models can now make a cheaper 480p draft when generate_video gets draft=true, and render_video_draft then turns an approved draft into a separate paid 1080p final. The README does not mention this flow yet. The instructions the server hands to clients also still pin an older package version in their npx command, so copy the command from the README instead. The project description still talks about ComfyUI and a 1,400+ prompt library. Both were removed in the 3.x line, and gallery search now always queries the live MeiGen API.

What it needs

To run it yourselfNode.js 22+
Runs onmacOS, Windows, Linux
MeiGenPaid planCloud generation, video and the visual workflows use purchased credits. Locally this needs an existing API key; new users connect the hosted endpoint with OAuth
OpenAI-compatible image APIPaid planOptional instead of MeiGen, for plain image generation only

Node.js 22 or newer is recommended for the npm package. Locally, set an existing MeiGen key in MEIGEN_API_TOKEN, or OPENAI_API_KEY with OPENAI_BASE_URL and OPENAI_MODEL for image generation on another provider. Keys can also sit in ~/.config/meigen/config.json. The hosted endpoint needs a client that does OAuth for remote MCP. In Codex run `codex mcp login meigen` after adding it, and in OpenClaw `openclaw mcp login meigen`.

Limits

  • Not covered The local server and API keys cannot reach the private library or saved characters. search_favorites, list_boards, search_history, save_favorite and list_characters exist only on the hosted OAuth endpoint
  • Limit The makers no longer hand out new MeiGen API keys for the local package. New users are sent to the hosted endpoint with OAuth
  • Not covered Video, video drafts and the five visual workflows run only on MeiGen Cloud and need MEIGEN_API_TOKEN, not an OpenAI-compatible provider
  • Cost Generation, drafts, finals and visual workflows spend purchased MeiGen credits. Daily free credits do not apply. A draft's 1080p final is charged in full on top of the draft
  • Size cap Locally, search_gallery returns 3 results by default and never more than 3 per call, although limit accepts 1 to 20
  • Limit Uploaded reference media gets public links that anyone holding the link can open. They are cleared seven days after upload
  • Limit Long video jobs can run 15 minutes or more, past the default tool timeout of many clients. SETUP.md suggests raising it
  • Limit The server's built-in instructions name an older pinned npx version than the one published. Use the command from the README

The makers no longer issue new MeiGen API keys for the local package: new users are sent to remote OAuth, and API keys and the local server cannot reach the private library or saved characters (list_characters is not in the local tool set). Characters can be read and reused but not created. Video and the five visual workflows run only on MeiGen Cloud, not on an OpenAI-compatible provider. Locally, search_gallery returns 3 results by default and at most 3 per call even though `limit` accepts up to 20. Uploaded reference media gets public links that anyone with the link can open, cleared after seven days. Long video jobs can run 15 minutes or more, beyond many clients' default tool timeouts.

Who it suits

Suits someone already using MeiGen who wants to search its prompt gallery and run paid image, video or product-photo jobs from an assistant; the hosted OAuth endpoint is the path for new users and the only one that sees a personal library. The local npm form suits existing key holders, or anyone who wants plain image generation against their own OpenAI-compatible endpoint plus local prompt rewriting. Anyone wanting free, offline or self-hosted generation (the old ComfyUI path) should look elsewhere, since 3.x calls MeiGen's or another provider's servers for every generation.

Tools 18

7 only read, 11 change things. A ? marks an optional parameter.

search_gallery
Search MeiGen's public gallery of example images and their prompts, with paging
get_inspiration
Fetch the full prompt and images for one gallery entry
enhance_prompt
Rewrite a short idea into a fuller image prompt, locally
list_models
List image and video models with their options, capabilities and prices
manage_preferences changes
Read or update the stored style and model preferences
generate_image changes
Generate images on MeiGen Cloud or a configured OpenAI-compatible provider
generate_video changes
Generate a video on MeiGen Cloud, optionally as a 480p draft or from frames and media referencesprompt: stringmodel?: stringmodelId?: stringdraft?: booleantier?: stringduration?: integerresolution?: stringaspectRatio?: stringfirstFrame?: stringlastFrame?: stringcontentReferenceImages?: arrayreferenceVideo?: stringreferenceVideos?: arrayreferenceAudios?: arrayreferenceVideoDuration?: integer
render_video_draft changes
Render an approved video draft as a separate paid 1080p finalrequestId: stringsourceDraftGenerationId: stringconfirmedCredits: numberretryFailedGenerationId?: string
check_generation
Check the status and results of an image or video job, including a draft's final price
list_skills
Describe the five visual workflows, their inputs and live pricesskill?: enum
upload_skill_image changes
Prepare a local or external image as a reference and return its URL
upload_reference_media changes
Prepare an MP4/MOV or WAV/MP3 clip as a video-model reference
remove_background changes
Cut the subject out of an image onto a transparent backgroundrequestId: stringproductImage: string
upscale_image changes
Enhance an original image in crisp or creative moderequestId: stringimageUrl: stringmode?: enumallowDownscale?: booleanconfirmedCredits: integer
generate_marketing_poster changes
Design a marketing poster from a briefrequestId: stringbrand: stringcontent?: stringextraNotes?: stringlogo?: stringproductImages?: arraystyleImage?: stringstyleId?: stringcustomStyle?: string
generate_product_detail_images changes
Make a set of product detail shots from a product photorequestId: stringproductImage: stringproductName?: stringmodules: arraycustomModules?: arrayplatform?: stringlanguage?: stringquality?: enumaspectRatio?: stringmodelImage?: stringlogo?: stringextraProductImages?: array
generate_ai_background changes
Place a product on a newly generated backgroundrequestId: stringproductImage: stringmode?: enumratio?: stringquality?: enum
check_skill
Read results, partial failures and refund state of a visual workflow job

Similar servers