Stability AI MCP. Generate, refine, and clean professional visual assets instantly.

Stability AI MCP Server connects your agent straight into powerhouse visual models. You don't have to jump between a bunch of different external programs just to handle images; you run the whole gig right here in your chat interface. When you need mockups, or if you gotta clean up some messy product photos, this thing handles it all conversationally.

Your agent talks directly to these tools. You tell it what you want—a specific style of image, a cropped background, or just higher resolution—and the server executes it. It manages everything from synthesizing brand new visuals based on detailed text prompts to fixing low-res assets and surgically removing backgrounds. Here's how you use the core functions:

Generating Images From Scratch: When you need an entirely new visual, you've got a few routes depending on what you want the final product to look like. You can hit up generate_sd35 if you wanna use Stable Diffusion 3.5; this lets you choose from three separate, massive models for different looks.

If speed and quality are your main concerns, fire up generate_core_v2. For assets that need to be absolutely perfect—like stuff going into final print production—use generate_ultra_v2; it's built for high-end photorealism. You can also use text_to_image_v1 with v1 engines, but remember you gotta specify the engine ID, along with your prompt, width, and height.

Improving Existing Photos: If a source image is fuzzy or too small, don't sweat it. Run upscale_image. This tool blows up the dimensions of an existing picture without losing structural detail—it keeps things sharp so there’s no pixelation headache. Want to change a photo's style or swap out a subject? You use image_to_image_v1.

It takes your original source image and transforms it based on a new text prompt, but you gotta provide both the engine ID and that source picture for it to work.

Pinpoint Editing: Sometimes you don't want to change the whole photo; you just need to tweak one specific corner. For that, inpaint_image is your move. It lets you target an isolated section within an image and guide the replacement with a focused text prompt. If you gotta clean up a product shot by getting rid of a messy backdrop, use remove_background.

This tool automatically finds and strips out complex backgrounds from any product or scene photo instantly.

Engine Management and Credits: Before running anything expensive, you should always check your balance with get_credit_balance to make sure you don't get hit with unexpected usage fees. If you’re using the v1 tools (text_to_image_v1, etc.), you might need to know which engines are available first; run list_engines to pull a list of all engine IDs needed for those v1 functions.