
MeltFlex now has an MCP connector, and it changes where interior design happens. Instead of opening our app, uploading a photo, and clicking through a flow, you can stay inside the tool you already work in, Claude, Cursor, Claude Desktop, Windsurf, Codex, VS Code, or any agent that speaks the protocol, describe the room you want, attach a photo, and get a photorealistic redesign back in the conversation. No second editor, no copy-pasting between tabs, no separate billing to set up.
We will be straight about the scope, because that is the whole point of this connector: it does one thing, and does it well. It is not a thirty-model everything-studio. It turns a room photo into a believable redesign, on your own credits, from inside any agent. This guide covers what it is, what it can do, how to connect it in about a minute, and where it actually fits.
The short version
generate_interior restyles a room photo; check_credits reports your balance. That is the whole surface.MCP, short for Model Context Protocol, is an open standard that gives AI agents secure access to external tools. Claude and most modern agents support it natively. MeltFlex ships a small MCP server, the meltflex-mcp package, that exposes our interior generation as tools any agent can call. Connect the two and your chat thread becomes a place where you can actually redesign rooms, not just talk about them.
Under the surface it is deliberately thin. The server is a client over the MeltFlex API, the same engine behind our app, so it inherits everything the API already enforces: your credits, your limits, your refunds. That is also why it is safe to hand an agent: the connector cannot do anything you could not already do with a single API request using the same key.

The kind of result that comes back in your chat: a photorealistic redesign of a real room, saved to disk as a file you can use.
The surface is small on purpose. There are exactly two tools, and you will mostly use one.
generate_interior (10 credits). Restyle a room photo into a photorealistic redesign. You give it a prompt (the style and pieces you want) and an image (a file path, URL, or data URL). Optionally you pass reference_images to anchor specific furniture or decor, and an output_path to choose where the result is written. It saves the finished image to disk and returns the path.check_credits (free). Reports your account balance and the cost of each operation, so the agent can tell you what a batch will spend before it spends it.The reference-image input is the quietly powerful part. You can hand it a photo of an actual sofa, a paint swatch, or a piece you already own, and the redesign will build around it instead of inventing furniture you can never find. That is the difference between a pretty picture and a plan you can act on, and it is the same capability that powers shoppable results in the MeltFlex app.
Setup takes about a minute. You add MeltFlex as an MCP server in your client, authenticate with your API key once, and start prompting. Here are the two most common clients; the full setup page covers the rest.
Claude Code is a single command, then a one-time sign-in:
claude mcp add meltflex -- npx -y meltflex-mcp
npx -y meltflex-mcp auth mf_sk_xxxxxxxxxxxxCursor, Claude Desktop, and Windsurf use the standard MCP config shape. Drop this into your client’s mcp.json (or claude_desktop_config.json), with your own key embedded so there is no separate sign-in:
{
"mcpServers": {
"meltflex": {
"command": "npx",
"args": ["-y", "meltflex-mcp"],
"env": { "MELTFLEX_API_KEY": "mf_sk_xxxxxxxxxxxx" }
}
}
}You generate your key in account settings (it looks like mf_sk_…). Once the server shows as connected, you just ask in plain language:
Here’s my living room /Users/me/room.jpg — redesign it in warm minimalist style with a low linen sofa, light oak, and a big jute rug. Bright daylight.
The agent calls generate_interior, the redesign renders, and the finished image lands on your disk with the path returned in chat. From there you iterate the way you iterate on anything in that thread: change the style, swap a piece, run another room.
Three pieces do the work. Your agent reads your intent and turns plain language into a structured generate_interior call, picking up context from the conversation so “now try it darker” means the right thing. MCP is the secure bridge that lets the agent call the tool, with your API key staying in your environment and the agent only ever seeing the result, never the credential. MeltFlex runs the actual generation, the same engine as our app and API, deducts the credits, saves the image, and hands the file path back.
From your side it looks like one prompt. Underneath, the credit is deducted before the render and automatically refunded if the generation fails, so a bad call never quietly costs you. Nothing about the billing or the limits moves to the client; it all stays server-side where it belongs.

The redesign comes back where you already work. No second app, no separate editor, no exporting between tools.
1. One redesign, in seconds. The simplest path. Point it at a photo, describe the look, and the redesign comes back in the same chat. Good for a single room, a quick “what would this look like”, or testing a style before you commit.
2. Anchor it to real furniture. Pass reference images of pieces you actually want, a specific sofa, your existing dining table, a paint colour, and the redesign builds around them instead of guessing. This is how you go from a mood board to something you can genuinely buy and place.
3. Wire it into your own work. Because the server is also a CLI and a thin API client, developers can script it. A real-estate team can batch-stage a folder of empty listing photos from Claude Code or the terminal; an app can shell out to the same package. One brief, many rooms, all on your own credits.

A single generate_interior call: the same room, before and after. The window and proportions stay put while the styling changes.
That is the whole setup. The connector runs against your existing MeltFlex credits, so there is nothing extra to install on the server side and no separate billing to negotiate.
Interior visualization has lived behind its own door for years: a separate app, a separate login, a separate place you go when you want to see a room differently. Every one of those context switches is friction, and friction is where ideas quietly die between “I wonder” and actually seeing it.
MCP removes the door. The moment your agent can redesign a room without you leaving the conversation, the cost of trying an idea drops to a sentence, and when trying is that cheap, you try more. That is the real win here, not raw speed but a short enough loop that you actually iterate, and iteration is where rooms get good. We kept the connector narrow precisely so that loop stays fast: one clear job, done well, wherever you already work.
Connect MeltFlex to your agent
What is MeltFlex MCP?
A connector that lets any Model Context Protocol agent — Claude, Cursor, Claude Desktop, Windsurf, Codex, VS Code — redesign room photos from chat. It runs as the meltflex-mcp server, a thin client over the MeltFlex API, and every generation is billed to your own credits.
Which clients work with it?
Any MCP-capable agent. We document one-step setup for Claude Code, Cursor, Claude Desktop, Windsurf, Codex, and VS Code, and the same package doubles as a command-line tool.
Do I need an API key, and whose credits are used?
Yes, your own key (from settings), and your own credits — 10 per redesigned image, refunded if a generation fails. Checking your balance is free.
Is it safe?
It adds no new attack surface. Your key stays in your environment, the agent only sees results, and all limits and refunds live server-side. Anything it can do is already doable with one API request using the same key.
Where do the generated images go?generate_interior saves the redesign to disk and returns the file path; pass output_path to choose the location. Nothing is trapped in the chat thread.