# Midjourney MCP > Midjourney MCP Server connects your AI agent to Midjourney's Imagine API, giving you direct control over visual output. Use it to generate images from text prompts, upscale low-resolution assets, create variations of concepts, or blend multiple source images into a single composition using the `blend` tool. It also lets you reverse-engineer prompts (`describe`) and edit specific areas in generated art with masking (`inpaint`). ## Overview - **Category:** ai-frontier - **Price:** Free - **Tags:** generative-ai, text-to-image, image-upscaling, visual-content, prompt-engineering, creative-automation ## Description You connect your AI agent straight into Midjourney's Imagine API via this server. That means you get direct, hands-on control over every single visual asset you create. Forget sending prompts out to a separate web interface; your agent handles the whole pipeline right here. **Generating Images from Text** To start something new, you run **`imagine`**. You feed it a text prompt and style modifiers, and it immediately spits back a task ID while kicking off the generation. It doesn't give you the image yet; it gives you the green light to check on it later. **Managing Tasks and Iterations** Once that job is running, you use **`get_task_status`**. This tool checks if the background task finished and tells you exactly what happened. When it's done, it hands you the resulting image URLs and any actions available next—like upscaling or making variations. If you need to see a list of everything your agent has generated recently, use **`get_tasks`**; this lets you find specific IDs needed for later upscale or variation calls. If the first batch of four images isn't hitting the mark, don't panic. You can run **`reroll`**. This tool takes your *exact* original prompt and generates another set of four brand new pictures. It’s like getting a second take on the same shot until you nail it. **Refining and Enhancing Artworks** If one of those initial images looks close but not perfect, you've got options for refinement. You can use **`variation`** to generate alternative versions of an existing image, exploring different artistic interpretations of that concept. If the resolution is too low for printing or high-detail work, run **`upscale`**. This takes any of the initial four thumbnails and blows it up into a full, high-quality file. For more granular control, you use **`inpaint`**. You don't change the whole picture; you mask off just one small area—say, someone’s hand or a piece of clothing—and then edit only that specific region. The rest of the art stays exactly where it is. **Advanced Composition and Analysis** Sometimes you need to combine multiple sources into one composition. You execute **`blend`** by providing several source image URLs, and optionally giving your agent a text prompt to guide how those images should mix together into a single new picture. For maximum creative control, you might also use the **`variation`** tool in conjunction with **`get_task_status`**, letting your agent cycle through options until it finds the perfect balance. If you're looking at an image and think, 'How did they even make this?', you run **`describe`**. This analyzes the uploaded picture and returns the structured text prompt—the actual words—that were used to create its style or composition. It’s total reverse-engineering power for your agent. You've got a full creative lifecycle here: from initial generation via **`imagine`**, through iterative refinement with **`variation`** and **`reroll`**, to high-fidelity adjustments using **`upscale`** and targeted edits with **`inpaint`**. You can even take disparate pieces of art and fuse them together using the **`blend`** tool, all while keeping track of everything via **`get_tasks`** and checking the status with **`get_task_status`**. It's one end-to-end system for your agent to build visuals from scratch, refine what exists, or reconstruct lost prompts. ## Tools ### blend Takes multiple image URLs and optionally a text prompt to combine them into a single new composition. ### describe Analyzes an uploaded image and returns the structured text prompt used to create it, useful for reverse-engineering styles. ### get_task_status Checks if an ongoing generation task is finished and provides the resulting image URLs and available actions (upscale/variation). ### get_tasks Lists recent image generation tasks, allowing you to find specific IDs for subsequent upscale or variation calls. ### imagine Generates a new set of images from scratch based on your text prompt and returns an immediate task ID. ### inpaint Edits only specific, masked regions within an existing generated image while keeping the rest of the art untouched. ### reroll Regenerates four brand new images using the exact same initial prompt if none of the first set were satisfactory. ### upscale Takes a specific thumbnail from a generation and increases its resolution to a high-quality file. ### variation Creates alternative versions of an existing generated image, exploring different interpretations of that concept. ## Prompt Examples **Prompt:** ``` Generate an image of a cyberpunk city at sunset, 16:9 aspect ratio. ``` **Response:** ``` 🎨 Image generation started! Task ID: mj_abc123. Check progress with get_task_status. Typical generation time: 30-60 seconds. ``` **Prompt:** ``` Upscale image position 2 from the last generation. ``` **Response:** ``` 🔍 Image upscaled successfully! Full resolution image URL: https://cdn.midjourney.com/... You can now download or share this image. ``` **Prompt:** ``` Describe this image and tell me what prompt would create it: https://example.com/art.jpg ``` **Response:** ``` 📝 Generated prompt: 'ethereal watercolor painting of a lone wolf howling at a blood moon, dark forest silhouette, mystical atmosphere --ar 16:9 --v 6'. Use this prompt with imagine to recreate a similar image. ``` ## Capabilities ### Generate Images from Text Run the `imagine` tool to create new images based on specific text descriptions and style modifiers. ### Enhance Resolution Use `upscale` to take any of the initial four generated thumbnails and return a full, high-resolution image file. ### Modify Specific Areas The `inpaint` tool allows you to edit only masked regions of an existing image without changing the rest of the composition. ### Combine Multiple Images Execute the `blend` tool by providing multiple source images and optionally a text prompt to guide their combination into one new picture. ### Deconstruct Artworks The `describe` tool analyzes an uploaded image and returns the detailed text prompt needed to recreate its style or composition. ## Use Cases ### Developing Character Concept Art A developer needs 10 different looks for a villain. They run `imagine` with the base prompt. The initial four images are okay, but not right. Instead of restarting, they use `variation` on image #3 to explore that specific angle, and then repeat the process until they nail the look. ### Creating a Composite Scene A marketing team has three reference photos: a background landscape, a character portrait, and an object. They run `blend`, providing all three URLs plus a prompt like 'dramatic lighting, cinematic shot.' The agent stitches them into one cohesive piece. ### Improving Source Material A designer finds a low-res sketch for a product mockup. They upload it and use `describe` to pull out the style keywords ('steampunk, brass accents'). Then they feed those keywords back into `imagine` to generate high-quality versions. ### Fixing Background Elements An agent generates a perfect portrait, but there's an unwanted power line in the background. Instead of regenerating the whole image, they use `inpaint`, masking just the wires and letting the AI fill in the missing detail naturally. ## Benefits - **Rapid Prototyping:** Don't manually adjust prompts. Use the `variation` tool to instantly explore multiple takes on a concept without rewriting text. You get immediate visual alternatives for every initial idea. - **Advanced Composition:** Need something complex? The `blend` tool lets you combine 2-5 source images, guiding the outcome with an optional prompt. This goes way beyond simple side-by-side collages. - **Image Quality Control:** Generating a thumbnail isn't enough. Use `upscale` immediately after generation to boost resolution on key assets before they leave your workflow. - **Iterative Workflow Management:** The system doesn't just generate and vanish. Use `get_tasks` to monitor all running jobs, ensuring you never lose track of a background image or asset ID. - **Smart Editing:** Forget Photoshop masks. With the `inpaint` tool, your agent edits specific sections—say, changing the color of a character's jacket—without touching anything else in the picture. ## How It Works The bottom line is: you talk to your agent, and the server handles all the API calls, status checks, and asset retrieval, giving you a finished image file in return. 1. First, subscribe to this server and input your Midjourney API key via Ace Data Cloud. 2. Second, issue a command (e.g., 'Generate an image of...') through your AI client, which calls the `imagine` tool. 3. Third, wait for the task ID; use `get_task_status` to check progress and retrieve the final high-resolution URL. ## Frequently Asked Questions **How do I know when my image generation task is ready using get_task_status?** The `get_task_status` tool reports the progress percentage and, critically, provides the final image URLs once the job is complete. If it's not finished, it gives you the current status instead. **Can I use blend to combine real-life photos with generated art?** Yes. The `blend` tool takes URLs of any images—whether they came from a photo library or were just created by Midjourney—and combines them using your guidance prompt. **What is the difference between reroll and variation in Midjourney MCP Server?** They serve different purposes. Use `reroll` when you want 4 completely new ideas based on a prompt. Use `variation` when you like one specific image but want to explore slightly different interpretations of just that concept. **What do I do if my generated art is low quality and needs fixing? Do I use inpaint or upscale?** If the whole image is too small, use `upscale`. If the overall resolution is fine but a specific part (like a hand or a background object) is wrong, mask that area and run `inpaint`. **If I need to track multiple ongoing generations, should I use get_tasks or check status with get_task_status?** Use `get_tasks` to list recent generation activity. This shows IDs and overall progress for everything you've run recently. You only use `get_task_status` if you know the specific task ID and need a precise, real-time update on its current percentage. **When using blend, are there restrictions on the source images I can provide?** The system allows blending between 2 to 5 image URLs. While you can mix different sources, remember that providing a guiding text prompt helps the AI know how to combine them into one cohesive composition. **What happens if I run `imagine` with an overly complex or ambiguous prompt?** The server will generate a task ID and thumbnail regardless, but the resulting image quality might be unpredictable. If you hit generation limits or receive an error, always check your API key credentials first. **How do I use describe to improve my prompt writing for future generations?** After using `describe` on a cool piece of art, the output gives you a text prompt. You copy that generated text and feed it directly into `imagine`. This is the fastest way to recreate a specific style or composition. **How do I get a Midjourney API key?** Sign up at [Ace Data Cloud](https://acedata.cloud/) and subscribe to the Midjourney Imagine API service. You'll receive an API key that works with this MCP server. **What's the difference between upscale and variation?** Upscale increases resolution of a specific grid image (1-4) to full quality. Variation creates 4 new creative alternatives based on that image's style and composition. Use upscale for final output, variation for exploration. **How long does image generation take?** Typically 30-60 seconds. Use get_task_status to check progress. The API returns a task ID immediately, and you can poll for completion or set a callback URL for async notification.