RELATED 01
MeiGen AI Design MCP
OFFICIAL SERVERImage and video generation, visual edit workflows and prompt-gallery search on MeiGen
1.8kGitHub stars
513/wknpm downloads
75popularity score
MCP Image Generator
Make, edit, upscale and vectorize images, and optionally videos, on your own GPU
By Erik (individual) Python Local · Streamable HTTP MediaAI and MLDesign
25 popularity score ★ 114
Connect your client
Local (once it is running)
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 --transport http image-generator http://localhost:5005/mcpCursor~/.cursor/mcp.json
{
"mcpServers": {
"image-generator": {
"url": "http://localhost:5005/mcp"
}
}
}VS Code.vscode/mcp.json
{
"servers": {
"image-generator": {
"type": "http",
"url": "http://localhost:5005/mcp"
}
}
}Codex~/.codex/config.toml
[mcp_servers.image-generator]
url = "http://localhost:5005/mcp"Gemini CLITerminal
gemini mcp add --scope user --transport http image-generator http://localhost:5005/mcpWhat it does
MCP Image Generator is a self-hosted server that makes and edits images on your own computer with the Qwen-Image-2.1 model. You can choose either a Turbo or a Non-Turbo model. Ask for an image in VS Code, Cursor, Claude Code, LM Studio or another MCP client and the result appears in the chat, with a link to the full file.
Its tools cover text to image, edits that combine up to 10 images (with an optional mask), photo restoration, transparent PNGs and background removal, seamless tiles, 360 panoramas, 2x or 4x upscaling, watermark removal and SVG vectorizing. An optional video extension adds 5 to 20 second videos with sound, transparent animations and GIFs, made with Kandinsky 6 Lite. Long jobs return a job_id that the client collects with get_job. A web gallery at port 5005 shows, searches and uploads images.
It runs in Docker. Start scripts for Windows, Linux and Mac detect the GPU, download 11-16 GB of model files and serve the MCP endpoint at http://localhost:5005/mcp (Streamable HTTP). It needs an NVIDIA GPU with 4 GB of VRAM or more (8 GB for video), or it can run in a slower CPU mode with 24 GB of RAM. The server code is MIT licensed, but most of the model files it downloads are for non-commercial use only.
Tools 15
generate_image- Text to image, with size, aspect ratio, transparent, tileable, seed, steps and negative prompt options
edit_image- Edit one image or combine up to 10, with an optional mask
generate_panorama- 360 panorama (2:1), optionally from a photo
remove_background- Cut out the subject into a transparent PNG
upscale_image- Enlarge 2x or 4x, up to 8192 px per side
remove_watermark- Remove watermarks, logos and overlaid text
vectorize_image- Convert a logo, icon or line art to an SVG file
get_job- Get the result of a job that was still running
cancel_job- Cancel a running job
list_images- List recent results and uploads
view_image- Let the model look at one image
server_status- Model, GPU placement, download progress and running jobs
generate_video- Video with sound from text or an image (video extension only)
extend_video- Make a longer video from a short one (video extension only)
video_to_gif- Turn a video into an animated GIF (video extension only)