This server bridges AI coding agents to a local InvokeAI creative engine, enabling full control over image generation, management, and system administration.
Image generation: Text-to-image (txt2img), image-to-image (img2img), inpainting (masked regions), and 4x RealESRGAN upscaling. Supports SD1.5, SDXL, Flux, and other model families. Configure parameters like prompt, negative prompt, model, seed, steps, dimensions (width/height), strength, CFG scale, and scheduler.
Queue management: Monitor queue status, list items, poll results, and download generated images. Cancel individual jobs, cancel batches, clear, pause, or resume the queue.
Gallery & image management: Browse, search by metadata, get details, download, star/unstar, and delete images.
Model management: List installed models by type, install from HuggingFace repo IDs, Civitai URLs, or local paths; update and delete models; track installation progress.
Board (collection) management: Create, list, get, update, and delete boards; add or remove images to/from boards.
Workflow management: List, view, save, and delete InvokeAI node workflows.
System introspection: Check InvokeAI health, version, configuration, cache statistics, and onboarding detection.
Dashboard cards: Display rich status cards for fleet, queue, models, and recent gallery images.
Utilities: Built-in help system with topic-based documentation; graceful server shutdown.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@invokeai-mcp Generate a photorealistic portrait of an astronaut riding a horse on Mars"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
A bridge between AI coding agents (Claude, Cursor, opencode) and your local InvokeAI creative engine - text-to-image, image-to-image, masked inpaint, and upscaling on your own GPU, plus full queue, model, gallery, board, and workflow management, with a polished dark webapp.
This repo wraps InvokeAI, the open-source professional Stable Diffusion /
Flux creative engine (Apache-2.0, 27k+ stars). InvokeAI runs as a local web
server (default http://127.0.0.1:9090) and is installed separately through
its launcher - it is never bundled here. See docs/WRAPPEE.md.
Related MCP server: ComfyUI MCP Server
Dashboard | Generate |
|
|
How it runs: headless bridge - this MCP server talks to your running InvokeAI instance over its REST API. InvokeAI must be installed and running (launcher install, model download, first run). Nothing is bundled.
Direction | Artifacts | Notes |
Hands-in | prompts, negative prompts, images (img2img/inpaint), mask images, model sources | Text or uploaded images |
Hands-out | generated PNGs, image URLs, local file paths, queue state |
|
Generate SD1.5 / SDXL / Flux / SD3.5 / Qwen Image images on your RTX 4090
8 generation modes with horizontal tabs: txt2img, img2img, inpaint, outpaint, upscale, ControlNet, IP-Adapter, seamless tiling
60-style catalog + 24 material presets with select-all batch generation and live per-item progress
AI prompt refiner (local LLM) that understands the selected style/material
Plugin surface: install custom node packs from git, browse 250+ built-in engine nodes
Transform existing images (img2img) and repair regions (masked inpaint)
4x RealESRGAN upscaling
Install models from HuggingFace or Civitai without leaving the chat
Full queue control: status, cancel, clear, resume, result polling
Gallery search, boards, star/favorite organization; sort/filter by style, painter, franchise, board, starred; batch ops (star/unstar/delete/zip/move) with exact per-image attribution
76 curated styles + 23 franchise presets + 60 painters + 1321 A1111 community style pack entries - all batchable in any combination
Workflow library management (save/load node workflows)
Dark SOTA webapp: Generate, Gallery, Models, Queue, Boards, Workflows, Inbox, Tools, Skills, Chat (local LLM), Settings, Help, Logs
Model | Family | VRAM (fp16) | License / gating | Notes |
FLUX.1 Dev / Schnell | flux | 12-16 GB (fp8 ~7 GB) | Gated (token) / schnell Apache-2.0 | Schnell: 4-step, structural quality |
FLUX.2 Dev / Klein 9B / 4B | flux2 | 18 GB / 9 GB / 4 GB | Gated (token) | Current flagship |
SD 3.5 Large / Medium | sd-3 | 16 GB / 5 GB | License accept + token | Rivals FLUX.1-dev |
SDXL + community (Juggernaut XL, RealVis) | sdxl | 8 GB | Mostly open | Default: Juggernaut XL v9 |
CogView4-6B | cogview4 | 12 GB (fp8 ~7 GB) | Apache-2.0, no gating | Great text rendering, bilingual |
Z-Image Turbo / Base | z-image | 12 GB (fp8 ~6 GB) | Permissive | Turbo is very fast |
Qwen-Image 20B | qwen-image | ~40 GB (fp8 ~20 GB) | Apache-2.0, no gating | Best prompt adherence; too heavy for 24 GB with desktop apps |
SD 1.5 / 2.x | sd-1 | 4 GB | Open | Legacy, fast iteration |
Krea 2, Ideogram 4, Anima, ERNIE | external | varies | Closed / API | Engine supports; not self-hostable |
Graph builders ship for sd-1, sdxl, flux, cogview4 (verified against engine 6.13.7);
sd-3, qwen-image, z-image, flux2 wiring is mechanical follow-up on request.
Models install to N:\InvokeAI-models (configurable via models_dir in invokeai.yaml).
The fastest path is the MCPB bundle for Claude Desktop:
Download invokeai-mcp-0.1.0.mcpb from
Releases
Open Claude Desktop and drag the file onto the window
Complete onboarding (install InvokeAI, download a model) - see docs/ONBOARDING.md
Other methods (mcpb CLI, manual config, webapp dev stack) are in INSTALL.md.
"Generate a neon cyberpunk city at night, rain, cinematic lighting"
"Use this image and make it a watercolor painting: [gallery image]"
"What models do I have installed? Install SDXL base from HuggingFace"
"Show me my recent images and download the last one to disk"
Doc | Contents |
All install methods, prerequisites | |
First-timer InvokeAI setup, model downloads, pitfalls | |
What InvokeAI is, official links, community | |
System architecture, graph builders, ports | |
Env vars, config options | |
All available tools | |
Contributing, local setup | |
Common issues |
InvokeAI installed and running (launcher install; models downloaded) - free, Apache-2.0
GPU with 6-12 GB VRAM recommended (SD1.5: 4 GB, SDXL: 8 GB, Flux: 12 GB+)
Engine install: invokeai-web manual install (uv venv + uv pip install invokeai)
or the official launcher; the engine serves its full canvas GUI on 9090
Claude Desktop (or any MCP client) for chat use; a browser for the webapp
Python 3.12+ and uv for source installs; Node/Bun for the webapp
MIT
Unclaimed servers have limited discoverability.
If you are the server author, to access and configure the admin panel.
Build and run visual creative-production workflows from your AI agent.
Generate images, video, music and voice from your CLI or AI agent. On-brand AI media toolkit.
Generate images, GIFs, and PDFs from HTML, URLs, or templates — from your AI agent.
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/sandraschi/invokeai-mcp'
If you have feedback or need assistance with the MCP directory API, please join our Discord server
