Standard Operating Procedure

Claude + Magnific MCP: generate production assets from chat

Connect Claude to the full Magnific platform via MCP and produce images, video, audio, 3D, and branded workflows without leaving your conversation.

20 min estimated setup
Intermediate
Updated June 2026
Tools needed Claude Desktop or Claude Code Magnific paid account magnific.com Chrome or any modern browser

Purpose

This SOP covers the complete setup and daily use of Magnific MCP inside Claude. Once connected, you can generate production-grade images, video clips with audio, product mockups, 3D assets, and voiceovers by describing what you need in plain language. No file exports, no tab switching, no manual prompt relay between tools.

This SOP is for founders and content operators who already use Claude and want to add a professional creative output layer without hiring a designer or running separate tools.

01

Create your Magnific account and confirm your paid plan

Magnific MCP is available on all paid plans. You need a live subscription before the connector will authenticate.

  1. aOpen a browser and go to magnific.com. Click Sign up in the top navigation.
  2. bComplete registration with your email. Verify the address from the confirmation email.
  3. cNavigate to Pricing and select a paid plan. The Standard plan covers all MCP tools. Higher tiers add more monthly credits.
  4. dComplete checkout. Your account is active as soon as payment is confirmed.
  5. eLog into your Magnific dashboard and confirm your credit balance is showing. This confirms the plan is active.
Warning

Every MCP action consumes credits, even on plans that include unlimited browser-based generations. Budget your credit usage accordingly when running MCP sessions at volume.

03

Add the Magnific connector to Claude

There are two paths depending on which Claude client you use.

Path A: Claude.ai web (Settings method)

  1. aOpen claude.ai and sign in to your account.
  2. bClick your profile icon in the top-right corner. Select Settings from the dropdown.
  3. cIn the left sidebar, click Connectors.
  4. dClick + Add connector. In the search field, type Magnific. If it appears in the directory, click it and then click Connect.
  5. eIf Magnific does not appear yet in the directory (it is listed as a custom connector while the official listing is in progress), click Add custom connector and paste the URL below.
URL https://mcp.magnific.com
  1. fName the connector Magnific and click Add connector.
  2. gAuthorize the connection when prompted. You may be redirected to Magnific to grant access. Log in with the same account from Step 1.

Path B: Claude Desktop (config file method)

  1. aOpen Claude Desktop. Go to Settings and locate the MCP configuration section.
  2. bOpen your claude_desktop_config.json file. Add the following block inside the mcpServers object:
JSON "magnific": { "url": "https://mcp.magnific.com", "transport": "http" }
  1. cSave the file. Restart Claude Desktop.
  2. dOn next launch, Claude will request Magnific authentication. Complete the browser flow with your Magnific credentials.
04

Verify the connection is live

Before generating anything, confirm Claude has access to the Magnific tools. This takes 30 seconds.

  1. aOpen a new Claude conversation.
  2. bType the verification prompt below and send it.
Prompt How many Magnific credits do I have available right now?
  1. cClaude should return your current credit balance. This call is free and does not consume credits. It confirms the connector is authenticated and working.
  2. dIf Claude says it does not have access to Magnific tools, return to Step 3 and confirm the connector is saved and authorized.
Tip

Checking your credit balance is always free. Use it at the start of any heavy session to make sure you have enough credits before running large batch jobs.

05

Generate your first asset from Claude chat

With the connection live, you can now direct Magnific's full creative suite through natural language. Claude handles model selection in auto mode. You can override it by naming a model.

Basic image generation:

Prompt Generate a cinematic product photo of a dark ceramic mug on a warm wooden table, morning light, shallow depth of field.

Specify a model explicitly:

Prompt Generate the same image using Google Imagen 4. Then generate a second version using Recraft V4 and show me both.

Upscale and reformat in one request:

Prompt Take the hero image from this generation, upscale it to 4K, then resize it for three formats: a 1:1 Instagram post, a 9:16 story, and a 16:9 YouTube thumbnail.

Video with audio in one request:

Prompt Generate a five-second cinematic video clip of waves at golden hour, add ambient ocean sound, use Kling 3.0 Omni.
Important

Every generation saves automatically to your Magnific history. You can search past generations from any session. Nothing is lost if you close the chat window.

06

Use Magnific Spaces to save repeatable workflows

Spaces let you save multi-step production flows as named workflows. Once saved, you can call the workflow by name from any Claude conversation and pass in only the variables that change.

  1. aDescribe the full workflow you want to save. Be specific: include every step, the models you want, the formats, and the naming convention.
Prompt Create a workflow called product-ad-pipeline that does the following: take any product photo I give it, generate a lifestyle version of the photo, upscale it to 4K, resize it for Instagram square, Instagram story, and YouTube thumbnail, then produce a five-second video clip of the lifestyle image with ambient background audio. Save this as a Space.
  1. bClaude will create the Space in Magnific and confirm it has been saved under the name you specified.
  2. cTo run the workflow in any future session, use:
Prompt Run product-ad-pipeline on this product photo. [attach or describe the image]
  1. dTo update a saved workflow, tell Claude what to change. Claude will update the Space in Magnific without requiring you to rebuild from scratch.
Tip

Name your Spaces using the same naming convention as your content types. Examples: carousel-asset-pipeline, ugc-ad-set, brand-mockup-pack. This makes them easier to call from chat without ambiguity.

07

Apply LoRAs for brand-consistent output

LoRAs are trained style, face, or product models. Once a LoRA is trained on your brand, you reference it by name in every generation to keep visual identity consistent across all assets.

  1. aGo to your Magnific dashboard. Navigate to the LoRAs section.
  2. bClick Train a LoRA. Choose the type: face (for a person), product (for an object or package), or style (for a visual treatment).
  3. cUpload 10 to 20 reference images. The cleaner and more consistent the reference set, the more accurate the LoRA output.
  4. dName the LoRA clearly. Example: brand-product-v1 or founder-face-v1.
  5. eWait for training to complete. Magnific will notify you when the LoRA is ready.
  6. fIn Claude, reference the LoRA by name in your generation prompt:
Prompt Generate a lifestyle product shot on a marble countertop, warm kitchen morning light, using the brand-product-v1 LoRA.
Note

LoRAs are compatible across most Magnific generation tools. You can apply a product LoRA to both still image generation and video clip generation in the same request.

08

Run batch and multi-format jobs from chat

Magnific MCP supports batch image generation, character creation across multiple scenes, and video concatenation. This is where the connector replaces the work of a full creative production round.

Batch image generation:

Prompt Generate 10 variations of this product in different lighting setups: morning light, studio white, golden hour, overcast outdoor, and low-key dark. Produce two versions of each.

Character across multiple scenes:

Prompt Use the founder-face-v1 LoRA to generate three scenes: the founder at a laptop in a coffee shop, the founder presenting at a whiteboard, and the founder in a one-on-one client meeting. Keep the face and style consistent across all three.

Run a saved workflow on a new asset:

Prompt Apply product-ad-pipeline to the new SKU image I'm attaching. Generate the full asset set.
Warning

Batch jobs consume credits at the rate of the individual operations. A 10-image batch with upscaling counts as 10 generation credits plus 10 upscale credits. Check your balance before running large batches.

Reference

Prompt templates for daily use

Copy and adapt these prompts for your most common creative production tasks.

Voiced video clip

Generate a cinematic shot of [subject] at [setting and time of day], give the subject a voice narrating [what they say], and produce a five-second clip.

Hero to social formats

Take the hero image from this generation, upscale it to 4K, and resize it for Instagram square, Instagram story, and YouTube thumbnail.

Save a new workflow

Save these steps as a workflow called "[workflow-name]" in my Magnific Spaces.

Run a saved workflow

Call "[workflow-name]" and run it on this [image / concept / brief].

Brand-consistent character sheet

Using the [lora-name] LoRA, generate [number] scenes showing [character] in [list of situations]. Keep face and style consistent across every shot.

Check and browse past work

Search my Magnific history for generations tagged with [keyword or date range] and show me the results here.

Reference

Available models

Magnific updates its model catalog daily. Below are the confirmed models as of June 2026. In auto mode, Claude chooses the best model for your request.

Model Type Best for
Nano Banana 2ImageFast concept generation, mood boards, variation exploration
Google Imagen 4ImageHigh-fidelity photorealistic output, product photography
Recraft V4ImageVector-friendly, brand-safe illustration and icon generation
GPT-image-2ImageInstruction-following, composite scenes, text-in-image
Seedream 5ImageCinematic stills, editorial and fashion aesthetics
Seedance 2.0VideoSmooth motion, product demos, ambient scene video
Kling 3.0 OmniVideo + AudioVideo with synchronized audio, character movement
Google Veo 3.1VideoHigh-resolution cinematic video output
Happy HorseVideoFast video generation, short-form content
Note

In auto mode, the MCP does not return which model it used. To confirm the model after generation, check the creation in your Magnific history dashboard. For full control, always name the model in your prompt.

FAQ

Frequently asked questions

Does MCP access cost extra on top of my Magnific plan?

No. MCP is available on all paid plans at no additional charge. Each operation within a session consumes credits from your plan allocation, including on plans that include unlimited browser-based generations.

Which Claude clients work with Magnific MCP?

Claude Desktop, Claude Code, and Claude.ai web all support MCP connectors. Magnific also works with ChatGPT, Cursor, VS Code, Codex, Gemini, and Hermes for teams running multi-tool setups.

Can I combine Magnific with other connectors in the same conversation?

Yes. You can use Magnific alongside other active connectors in the same session. For example, generate an image via Magnific and immediately send it as a Slack message or attach it to a Gmail draft, within the same Claude conversation.

Where do my generated assets go?

Every generation saves automatically to your Magnific account history. You can search, organize into projects, and share assets directly from your Magnific dashboard. Nothing is tied to a specific Claude conversation.

How does auto model selection work?

The MCP server reads your prompt and selects the most appropriate available model based on context. It does not return the model name in the response. To force a specific model, name it explicitly in your prompt.

Is there a way to do folder-level upscaling?

Folder-level upscale is listed as coming soon as of the date of this SOP. Individual and batch upscaling are available now.

Troubleshooting

Common issues and fixes

Claude says it does not have access to Magnific tools

Return to Settings and confirm the Magnific connector is saved and shows a connected status. If it shows disconnected, re-authorize by clicking the connector and completing the login flow again. On Desktop, confirm the config file is saved and Claude has been restarted.

Generation fails or returns an error

Check your credit balance first using the verification prompt from Step 4. A failed generation with zero credits remaining is the most common cause. If credits are available, try naming a specific model in your prompt to bypass auto mode and isolate the issue.

LoRA output does not match my reference images

Retrain the LoRA with a larger, more consistent reference set. Aim for at least 15 to 20 images with similar lighting, framing, and background conditions. Avoid reference images that mix multiple environments or lighting setups.

Saved workflow (Space) does not run as expected

Ask Claude to display the current contents of the Space by name. Review each step in the workflow and identify where the output diverges from your expectation. Tell Claude specifically what to change. Claude will update the Space directly without requiring a full rebuild.

Custom connector URL not accepted in Claude settings

Confirm the URL is entered exactly as https://mcp.magnific.com with no trailing slash. If the field still rejects it, try the one-click connector link from the Magnific MCP page, which pre-fills the URL and connector name automatically.

Document Record

Version history

Version Date Changes Author
1.0 June 2026 Initial release covering full setup, Spaces, LoRAs, batch jobs, model reference, FAQ, and troubleshooting AI Savvy Founders