Magnific runs a hosted MCP server that lets Claude upscale your photos 2x, 4x, 8x or 16x without leaving the chat. See how to connect it in two minutes, which prompts work, how Creative and Precision V2 differ, what each upscale costs, and which Picasso IA upscalers to compare.
You have a 900 pixel photo that looks fine on a phone and falls apart the moment someone zooms in. The usual routine means exporting the file, opening an upscaler in a browser tab, uploading it, waiting, and downloading the result. With the Magnific MCP server, Claude runs that step for you inside the chat. You connect once, ask for a 4x upscale in plain English, and the sharper file comes back in the same conversation. This article shows how the connection works, what to type, what it costs, when the REST API is the better tool, and which browser-based upscalers on Picasso IA are worth comparing against it.
MCP stands for Model Context Protocol, an open standard that lets an AI assistant call outside tools. Magnific runs its server remotely, so there is nothing to compile, no Docker container and no local process to keep alive. Claude talks to it over the internet, and the upscaler becomes one more thing Claude can do on request.
Here is what the setup looks like from your side:
https://mcp.magnific.comThe client keeps the session after the first sign-in, so you do not repeat the login every time you open a new chat.
Upscaling is only one slice. The server exposes a wider toolbox, and Claude picks the right tool from your request.
| Tool group | What it does |
|---|---|
| Account | Balance queries and project usage reports |
| Images | Generation, SVG output, upscaling, smart cropping, resizing, background removal, model browsing |
| Video | Video generation and a model catalog |
| Audio | Text to speech with voice selection |
| 3D | 3D model generation |
| Custom references | Character and style training |
| Library | Folders, Spaces, search and retrieval of past creations |
💡 Because everything lives in one chat, you can chain steps: generate an image, upscale it 4x, crop it to a banner ratio and remove the background, all without leaving the conversation.
The connection is a one-time job. Pick the route that matches how you use Claude.
Magnific and the URL https://mcp.magnific.com.Once the connector shows as active, ask Claude to check your Magnific balance. A correct number back means the whole chain works.
One command registers the server:
claude mcp add --transport http magnific https://mcp.magnific.com
Open /mcp inside a session to finish the browser sign-in. Cursor, Windsurf and VS Code follow the same pattern: add a remote HTTP server, paste the URL, authorize in the browser. ChatGPT works too, through its developer mode apps screen.
Why OAuth beats a pasted credential: nothing secret sits in a config file, nothing lands in your shell history, and you can cut access from your account page without hunting through every machine where you saved a token.
With the connector active, the request is a normal sentence. Attach a photo and write something like this:
Upscale this photo 4x with Magnific. Keep it faithful, do not invent new details, and tell me the credit cost before you run it.
💡 The exact settings Claude can pass depend on the tool schema in your session. Ask Claude to list them. If it cannot read an attached file, paste an HTTPS link to the image instead, since Magnific's own API accepts an image URL or a base64 string.
The creative upscaler offers 2x, 4x, 8x and 16x. The right factor depends on where the image ends up. A 1280×720 source at 4x becomes 5120×2880, which prints at roughly 17 by 9.6 inches at 300 dots per inch. The same source at 2x gives 2560×1440, enough for a sharp full screen display but only about 8.5 by 4.8 inches in print.
| Input size | Factor | Output size | Magnific's published example price |
|---|---|---|---|
| 640×480 | 2x | 1280×960 | €0.10 |
| 640×480 | 4x | 2560×1920 | €0.20 |
| 640×480 | 8x | 5120×3840 | €0.50 |
| 1280×720 | 2x | 2560×1440 | €0.10 |
| 1280×720 | 4x | 5120×2880 | €0.40 |
| 1920×1080 | 2x | 3840×2160 | €0.20 |
These figures come from the examples on the creative upscaler's documentation page. Prices can change, so confirm with the pricing page or ask Claude for a quote before a big run.
Magnific's API exposes two upscaler families, and the choice matters more than the scale factor. Which of them the MCP tool named "Magnific upscaler" uses for a given request is not spelled out in the docs, so test a sample image before you trust a whole batch.
The Creative upscaler adds or infers new detail, steered by an optional text prompt. It is built for stylized, higher impact results. Its controls include:
Resemblance pulls the result back toward your original, while creativity pushes it away. If a face drifts, raise one and lower the other.
Precision V2 is the faithful option. It adds three dials the first Precision release did not have.
| Setting | Range | Default | What it does |
|---|---|---|---|
| Sharpen | 0 to 100 | 7 | Edge definition |
| Smart grain | 0 to 100 | 7 | Keeps natural texture |
| Ultra detail | 0 to 100 | 30 | Micro-detail |
| Flavor | 3 options | none | sublime, photo or photo_denoiser |
| Scale factor | 2 to 16 | none | Output multiplier |
💡 For product photos, packaging, interface screenshots and anything where accuracy matters, start with Precision. Save Creative for tiny, soft or low quality sources where invented detail is a feature.
The REST API makes sense once upscaling becomes a pipeline instead of a chat. The creative endpoint is POST /v1/ai/image-upscaler and Precision V2 is POST /v1/ai/image-upscaler-precision-v2. Jobs are asynchronous: you create a task, then fetch it by ID or pass a webhook_url and let Magnific notify you when the output is ready.
| Question | MCP in Claude | REST API |
|---|---|---|
| Sign-in | OAuth in the browser | API credential in a request header |
| Best for | One-off jobs, small batches, mixed tasks | Scheduled jobs, apps, large queues |
| Control | Whatever the tool schema exposes | The full parameter list |
| Waiting | Claude handles it | You poll or use a webhook |
| Billing | Your Magnific credit balance | Credit-based, priced by output pixels |
Choose the MCP for anything you would do by hand twice a week. Choose the API when a script, an app or a nightly job needs the result without a person in the loop.
Both routes bill through Magnific credits. Costs vary by model and resolution, and the docs do not publish a cap on MCP usage.
The price depends on the area of the finished image, which comes from the input size and the scale factor together. It does not rise in a straight line. In the published examples, a 640×480 file costs €0.10 at 2x and €0.50 at 8x: sixteen times the pixels for five times the price. A 1280×720 file goes from €0.10 at 2x to €0.40 at 4x.
Here is a worked example. A shop owner has a 1280×720 product photo and needs a zoomable image for a store page. A 2x run produces 2560×1440 for about €0.10 at the published rate, which is plenty for a zoom viewer. The same owner also wants a poster of the hero shot, so only that single image gets the 4x run at about €0.40. Splitting the job by purpose keeps the bill close to the minimum.
Magnific is not the only option, and a second opinion on the same file is cheap. Picasso IA lists a full super-resolution category, so you can run one image through several upscalers and keep the winner. It also hosts large language models, including Claude Sonnet 5 and Claude Fable 5, which are handy for turning rough notes about a photo into a short prompt for a creative upscale.
| Model | Use it when |
|---|---|
| Topaz Labs Image Upscale | You want photographic detail from a name known on the desktop |
| Clarity Pro Upscaler | A soft source needs believable extra detail |
| Recraft Crisp Upscale | You need clean edges on graphics and product shots |
| Recraft Creative Upscale | The image can take some reinterpretation |
| Real-ESRGAN | You want a fast, widely used 2x or 4x pass |
| Crystal Upscaler | You are comparing another high fidelity option |
| Bria Increase Resolution | You want a plain resolution bump |
| Google Upscaler | You want a second opinion from a different engine |
| P Image Upscale | Speed matters more than fine control |
Upscalers work best on a clean source. If you are starting from nothing, generate the base image first with Picasso IA Image, Seedream 5 Pro or GPT Image 2. Fix flaws before you enlarge them, because an upscaler magnifies every smudge. Picasso IA Image Editor Pro handles that cleanup step.
💡 Always compare the same crop from each result. A full-frame view hides the differences that matter, such as waxy skin, smeared foliage and warped lettering.
Check that the connector is switched on for the current chat in Claude's tools menu. If it is, remove the connector and add it again to force a fresh OAuth sign-in. In Claude Code, run claude mcp list to confirm the server shows as connected, then /mcp to reauthorize. An expired session is the most common cause.
Both come from too much creative freedom. Pores and fine wrinkles, like those in the portrait above, are exactly what an over-eager upscaler smooths away. Try these in order:
If credits run out halfway through a batch, ask Claude for the balance first and split the job into smaller runs so a failure costs you one image, not fifty.
Pick one image you have always wanted bigger: a family portrait, a product shot, a landscape that never printed well. Connect the Magnific MCP, run a 2x test, then send the same file through Topaz Labs Image Upscale on Picasso IA and put both results side by side at 100% zoom. Two minutes of comparing will tell you more than any spec sheet.
Need a fresh image to enlarge? Generate it with Seedream 5 Pro, polish it in Picasso IA Image Editor Pro, and upscale it with whichever tool gives you the cleanest edges. Every model mentioned here, plus hundreds more, is waiting at picassoia.com/en/all-models. Open it, upload your photo, and see how much detail a small file can really hold.
