AI HomeDesign MCP

What AI HomeDesign MCP does

AI HomeDesign MCP MCP server exposes the AI HomeDesign V3 API as MCP tools for editing real-estate photos. It supports furnishing empty rooms, replacing furniture in furnished rooms, redesigning interiors, improving image quality, removing objects, converting exterior daylight to dusk, and changing wall, floor, or ceiling finishes.

The server is suited to workflows where an AI client receives a property image and needs to select an editing operation from a natural-language request. It does not provide a separate image editor interface; the MCP client discovers the tools and supplies their inputs.

How it works

An MCP client connects to the server over streamable HTTP at the /mcp endpoint. The server forwards the requested operation to the AI HomeDesign V3 API and returns URLs for the generated images. Each helper tool accepts either an image URL or base64-encoded image data.

AI HomeDesign MCP MCP server can use a caller’s AI HomeDesign API key for each request. The hosted connector creates an opaque token after validating the key, while self-hosted deployments can use a default API key from the environment. The implementation uses FastMCP with Starlette and uvicorn. PostgreSQL is optional for token mapping and usage logs.

Setup and configuration

The hosted option requires an AI HomeDesign API key. Generate a connector URL through the AI HomeDesign MCP page, then add that URL as a custom connector in Claude or another MCP-compatible client. The generated URL grants access to the associated account and should be treated as sensitive; rotating the API key revokes previously issued links.

For self-hosting, the repository supports Docker and Python. A Docker deployment listens on port 8080 by default and serves MCP at http://localhost:8080/mcp. A local Python run uses the project requirements and app/server.py.

The main environment variables are:

• AIHD_API_KEY: default or fallback AI HomeDesign API key.
• AIHD_API_BASE: API base URL, defaulting to the AI HomeDesign V3 endpoint.
• PORT: listening port, defaulting to 8080.
• DB_DSN and TOKEN_SECRET: optional values for database-backed token storage and encrypted key storage.
• MCP_CONNECTOR_BASE, UPLOAD_DIR, and PUBLIC_BASE: optional connector and upload configuration.

Tools and capabilities

AI HomeDesign MCP MCP server provides discovery tools named list_capabilities, list_styles, and describe_tool. These expose available operations, supported styles and room types, and example payloads.

Editing tools include:

• virtual_staging for furnishing an empty room.
• virtual_restaging for replacing furniture in an already-furnished room.
• interior_design for a complete redesign of a furnished room.
• enhance_image for lighting, sharpness, and color improvements.
• remove_items for furniture and clutter removal.
• day_to_dusk for changing a daytime exterior to dusk.
• change_wall, change_floor, and change_ceiling for surface restyling.

Supported styles include prime, modern, farmhouse, scandinavian, hampton, industrial, traditional, and contemporary. Supported room types include bedroom, living-room, kitchen, bathroom, dining-room, home-office, outdoor, and nursery. Lower-level project and process tools are also available for callers that need more control.

Limitations and notes

Image processing depends on the AI HomeDesign API and an applicable API key; the MCP repository itself is MIT-licensed, but API usage is separate. The server is stateless for image-editing requests, while optional database features support connector token storage and usage logging.

The connector token is not the raw API key, but possession of the token permits account access. The health endpoint does not require a key. If the optional database is unavailable, the server is designed to continue serving requests, but database-backed token and usage features may not be available.