Pedra MCP Server

Official Model Context Protocol server for the Pedra API — use Pedra's AI real-estate photo editing (virtual staging, renovation, room emptying, enhancement, sky replacement, object removal/blur, and property videos) directly from Claude, ChatGPT, Cursor, and any other MCP client.

It exposes one tool per API endpoint. Each tool is a single blocking call that returns the final asset URL(s) — there are no job IDs to poll.

Quick start

You need a Pedra API key — get one from your Pedra account. The server reads it from the PEDRA_API_KEY environment variable.

The server runs over stdio and is published to npm, so most clients just run it with npx — no global install needed.

Claude Desktop

Add this to your claude_desktop_config.json (Settings → Developer → Edit Config):

{
  "mcpServers": {
    "pedra": {
      "command": "npx",
      "args": ["-y", "@pedra-ai/mcp"],
      "env": { "PEDRA_API_KEY": "your-api-key" }
    }
  }
}

Cursor

In ~/.cursor/mcp.json (or .cursor/mcp.json in a project):

{
  "mcpServers": {
    "pedra": {
      "command": "npx",
      "args": ["-y", "@pedra-ai/mcp"],
      "env": { "PEDRA_API_KEY": "your-api-key" }
    }
  }
}

Any MCP client

Run the binary directly with the key in the environment:

PEDRA_API_KEY=your-api-key npx -y @pedra-ai/mcp

Or install it:

npm install -g @pedra-ai/mcp
PEDRA_API_KEY=your-api-key pedra-mcp

Related MCP server: pruna-mcp-server

Tools

Most image tools take an imageUrl (a public URL or a data: URL) plus a few optional parameters; see each tool's input schema in your MCP client. Example prompts once connected:

"Use Pedra to virtually stage https://example.com/empty-living-room.jpg as a minimalist living room."

"How many Pedra credits do I have left?"

How it works

This server is a thin wrapper over @pedra-ai/sdk, which encodes the API's contract details:

• Synchronous by design. Every endpoint blocks and returns the final URL(s) in the response body. Even pedra_create_video polls server-side and returns the finished videoUrl inline (it can take up to ~10 minutes; the API keeps the connection alive with a heartbeat).

• Errors are tool errors. The API's 4xx responses (insufficient credits, bad image, …) come back as MCP tool errors with a readable message, not crashes.

License

MIT © Pedra