Luw.ai

Official Luw.ai MCP server

AI interior, exterior and landscape design, photoreal rendering, image editing, video and 3D,

inside Claude, Cursor, VS Code, Windsurf, Codex and any other MCP client.

You:     Redesign ~/Desktop/living-room.jpg in Japandi style, give me 2 options.
Claude:  ⟶ luw_interior_design  ✓ 2 designs in 21s   [shows both images]
You:     Love the first one. Make the sofa green velvet, then turn it into a fly-through video.
Claude:  ⟶ luw_edit_image ⟶ luw_generate_video   ✓

Quick start

Add https://mcp.luw.ai/mcp to your client and sign in to Luw.ai. Nothing to install, no key to copy: signing in creates one for you. New accounts get free credits. To disconnect, delete that key on the API page.

Claude (claude.ai, Claude Desktop, mobile)

Settings → Connectors → Add custom connector, paste https://mcp.luw.ai/mcp, then Connect and sign in to Luw.ai.

Claude Code

claude mcp add --transport http luw https://mcp.luw.ai/mcp --scope user

Then type /mcp, pick luw and sign in.

ChatGPT

Settings → Apps & Connectors → Advanced settings, turn on Developer mode, then Create: URL https://mcp.luw.ai/mcp, authentication OAuth, and sign in.

Cursor

Click Add to Cursor above, or add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "luw": { "url": "https://mcp.luw.ai/mcp" }
  }
}

Cursor shows Needs login next to luw; click it to sign in.

VS Code (Copilot)

Click Install in VS Code above, or add to .vscode/mcp.json:

{
  "servers": {
    "luw": { "type": "http", "url": "https://mcp.luw.ai/mcp" }
  }
}

VS Code asks you to sign in the first time a tool runs.

Any other client with remote MCP support

For a client that can't sign in, create a key at app.luw.ai/dashboard/api and send it as a header:

{
  "mcpServers": {
    "luw": {
      "url": "https://mcp.luw.ai/mcp",
      "headers": { "Authorization": "Bearer YOUR_LUW_API_KEY" }
    }
  }
}

If it can neither sign in nor send headers, use https://mcp.luw.ai/mcp?api_key=YOUR_LUW_API_KEY. Anyone who has that URL can spend your credits, so keep it private.

The hosted server is stateless and stores nothing; your key goes straight to the Luw.ai API on each request. It can't read files from your computer, so give it image URLs, or use the local setup below.

Local: work with files on your computer

The local server runs on your machine with npx, so it can read paths like ~/Desktop/room.jpg. It needs Node.js 18+ and an API key from app.luw.ai/dashboard/api.

Claude Code

claude mcp add luw --scope user --env LUW_API_KEY=YOUR_LUW_API_KEY -- npx -y @luw-ai/mcp

Or install it as a plugin. Claude Code asks for your key and keeps it in secure storage:

claude plugin marketplace add Luvi-io/luw-mcp
claude plugin install luw@luw-ai

Claude Desktop

One click: download **luw.mcpb**, double-click it, paste your API key. No Node.js needed.

Or add this to claude_desktop_config.json (Settings → Developer → Edit Config):

{
  "mcpServers": {
    "luw": {
      "command": "npx",
      "args": ["-y", "@luw-ai/mcp"],
      "env": { "LUW_API_KEY": "YOUR_LUW_API_KEY" }
    }
  }
}

Cursor

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "luw": {
      "command": "npx",
      "args": ["-y", "@luw-ai/mcp"],
      "env": { "LUW_API_KEY": "YOUR_LUW_API_KEY" }
    }
  }
}

VS Code (Copilot)

Add to .vscode/mcp.json. VS Code asks for your key and keeps it in its secret storage:

{
  "inputs": [{ "type": "promptString", "id": "luw_api_key", "description": "Luw.ai API key", "password": true }],
  "servers": {
    "luw": {
      "command": "npx",
      "args": ["-y", "@luw-ai/mcp"],
      "env": { "LUW_API_KEY": "${input:luw_api_key}" }
    }
  }
}

Windsurf (Devin Desktop)

Add to ~/.config/devin/mcp_config.json (Windows: %APPDATA%\devin\mcp_config.json). Older Windsurf versions use ~/.codeium/windsurf/mcp_config.json.

{
  "mcpServers": {
    "luw": {
      "command": "npx",
      "args": ["-y", "@luw-ai/mcp"],
      "env": { "LUW_API_KEY": "YOUR_LUW_API_KEY" }
    }
  }
}

OpenAI Codex CLI

codex mcp add luw --env LUW_API_KEY=YOUR_LUW_API_KEY -- npx -y @luw-ai/mcp

Or in ~/.codex/config.toml:

[mcp_servers.luw]
command = "npx"
args = ["-y", "@luw-ai/mcp"]
env = { LUW_API_KEY = "YOUR_LUW_API_KEY" }

Gemini CLI

gemini extensions install https://github.com/Luvi-io/luw-mcp

Gemini CLI asks for your key and keeps it in the system keychain. Or add to ~/.gemini/settings.json:

{
  "mcpServers": {
    "luw": {
      "command": "npx",
      "args": ["-y", "@luw-ai/mcp"],
      "env": { "LUW_API_KEY": "YOUR_LUW_API_KEY" }
    }
  }
}

Kiro

Click Add to Kiro above, confirm, then replace YOUR_LUW_API_KEY in Kiro's MCP config.

Windows: if npx isn't found, use "command": "cmd", "args": ["/c", "npx", "-y", "@luw-ai/mcp"].

What you can ask

• "Redesign this bedroom (~/Downloads/bedroom.jpg) as Mid-Century Modern, 3 variations."
• "This listing photo is an empty room. Stage it as a cozy Scandinavian living room."
• "Render my sketch plan-sketch.png as a photoreal modern villa at golden hour."
• "Change the floor in this kitchen to herringbone oak." (segments the floor, then swaps the material)
• "Remove all the furniture from this room."
• "Put this chair photo on a terrazzo floor in a sunlit gallery for an ad."
• "Turn this render into a drone fly-through video."
• "Make a 3D model of this armchair from these 3 photos."
• "Seamless blue zellige tile texture, 1024px."
• "Ask ArchiGPT how to improve the feng shui of this floor plan."

Built-in prompts show up as slash commands or prompt templates in your client: redesign_room, stage_empty_room, sketch_to_render, product_shot.

Tools

Every generation uses your Luw.ai credits. Variations are billed one generation each.

Luw Studio: edit your 3D projects

Luw Studio designs buildings in 3D. With the studio tools, your assistant edits your Studio projects directly: rooms, walls, doors and windows, stairs, roofs, furniture and finishes. Each edit is saved to the project as it is made, and a Studio tab that has the project open shows it within seconds.

• Hosted: add https://mcp.luw.ai/studio instead of /mcp (same sign-in). It has every hosted tool plus the studio tools.
• Local: add studio to LUW_TOOLSETS, e.g. LUW_TOOLSETS=generate,archigpt,personas,projects,studio.

"In my Luw Studio project Beach House, add a 3×4 m bedroom east of the living room with a window, and furnish it."

The scene tools are Pascal's open-source MCP tools (MIT), running on the same core as Luw Studio, with its catalog (Luw's own pieces included) and its defaults. They don't use credits. A project last saved by the previous Luw Studio is readable; open it once in Studio to edit it here. luw://studio/guide is the building guide the assistant reads first.

Resources

Read-only context you can attach to a conversation. In Claude Code, type @luw: to pick one; other clients that support resources offer them as attachments.

How it works

• Local files just work. Any image argument accepts an https:// URL, a local path (~/Desktop/room.jpg), or a data: URI. Local files are uploaded to Luw.ai storage as temporary files (deleted after 12 hours) and cached for the session, so repeated edits don't upload the same file again.
• You see the results. Finished images are embedded in the tool result, so Claude and other clients show them inline. Set LUW_OUTPUT_DIR to also save every result to disk.
• Long jobs don't time out. A call waits up to 50 seconds and streams progress to clients that show it. If a job like a video takes longer, the tool returns a processing_url and the assistant collects it with luw_get_result, without generating (or paying) twice.
• Never billed twice. Every generation request carries an Idempotency-Key, so if a network error forces a retry, Luw.ai returns the original job instead of charging again.
• Masks are handled for you. luw_segment returns masks as URLs, so "change the floor" becomes segment ⟶ magic wand with no manual masking.
• Fast startup. The package is one bundled file with zero dependencies, so npx downloads a single ~0.9 MB package and starts right away.

Configuration

Self-hosting the remote server

The same package serves the hosted Streamable HTTP endpoint:

npx -y @luw-ai/mcp --http --port 8080 --host 0.0.0.0   # MCP at /mcp, health at /health

docker build -t luw-mcp . && docker run -p 8080:8080 luw-mcp

It's stateless. Each request brings its own key in Authorization: Bearer …, X-Luw-Api-Key, or ?api_key=. Set LUW_MCP_OAUTH_SECRET to also let clients sign in to Luw.ai instead (OAuth); the sign-in hands the client an ordinary API key. It deploys to Vercel as-is (vercel.json is included), and a Procfile is included for Heroku. See docs/maintainers.md for deployment and release steps.

Development

npm ci
npm test               # unit and protocol tests (no network)
npm run build          # bundles dist/cli.js
npm run inspector      # try every tool in the MCP Inspector
LUW_API_KEY=... npm run smoke            # live checks against the real API (free)
LUW_API_KEY=... npm run smoke -- --paid  # plus one real generation (1 credit)
npm run mcpb           # build/luw.mcpb, the Claude Desktop extension
node scripts/vendor-pascal.mjs ../luwstudio2  # rebuild vendor/: Pascal core + MCP as Luw Studio runs them

Support

• API reference: luw-ai.gitbook.io/api
• Issues: GitHub Issues
• Email: support@luvi.io

MIT © Luvi Technologies, Inc.