MCP Integrations

Aurora works as a Model Context Protocol (MCP) server, which means you can build and deploy projects directly from your favorite AI tools without ever opening the Aurora website.

What you can do via MCP

From any connected tool, you can:

Create new projects from a prompt
Turn something you built with your AI assistant into a real app with a backend and a live URL
List and open existing projects
Read, write, and edit project files
Deploy to production
Check deployment status

All through natural conversation with your AI assistant.

Supported tools

Claude

Anthropic’s AI assistant on claude.ai. Add Aurora as a connected integration.

Claude Code

Anthropic’s CLI tool for developers. Connect Aurora with one command.

Cursor

AI code editor. Install Aurora with one click.

VS Code

Microsoft’s code editor with GitHub Copilot. Install Aurora with one click.

LM Studio

Run local models. Install Aurora with one click.

Windsurf

Codeium’s AI editor. Connect in the MCP settings.

Before you connect

All tools use the same server URL:

https://mcp.aurora.build/mcp

Aurora uses OAuth discovery. You do not need an API key. The tool opens your browser, and you click Authorize.

Create your Aurora account at aurora.build before you connect a tool. Some tools, for example LM Studio, stop waiting for the authorization after about two minutes. If you sign up during the connection, the tool can stop waiting before you finish.

Connect your tool

Claude (claude.ai)

Add the connector

In Claude, go to Settings > Connectors. Click Add custom connector and paste the Aurora server URL.

Authorize

Claude opens Aurora in your browser. Click Authorize.

Build something

Ask Claude for what you want, for example: “Build me a landing page for my bakery.” Claude can show you a first version in the chat.

Put it into Aurora

Ask: “Put this into Aurora.” Claude creates an Aurora project from it and gives you a live preview link. From there, the app has a backend, its own GitHub repo, and a public URL when you publish.

Claude Code (CLI)

Add the server

Run this command:

claude mcp add aurora --transport http https://mcp.aurora.build/mcp

Authorize

Run /mcp in Claude Code, select aurora, and authenticate in your browser.

Start building

Ask for a project in your Claude Code session:

Create a new Aurora project called "my-dashboard" with a weather widget and dark mode

Cursor

Click the button. Cursor opens and adds the Aurora server. The button needs the Cursor desktop app.

Add Aurora to Cursor manually

Go to Cursor Settings > MCP, or open .cursor/mcp.json in your project.

Add Aurora to your configuration:

{
  "mcpServers": {
    "aurora": {
      "url": "https://mcp.aurora.build/mcp"
    }
  }
}

When you first use an Aurora tool, Cursor opens your browser. Click Authorize.

VS Code

Click the button. VS Code opens and shows the Aurora server. Click Install.

Add Aurora to VS Code manually

Open the Command Palette and run MCP: Open User Configuration. You can also edit .vscode/mcp.json in your workspace.

Add Aurora to your configuration:

{
  "servers": {
    "aurora": {
      "type": "http",
      "url": "https://mcp.aurora.build/mcp"
    }
  }
}

Start the server. VS Code opens your browser. Click Authorize. Then open Copilot Chat in Agent mode and ask it to create an Aurora project.

LM Studio

Sign in to aurora.build in your browser before you start. LM Studio stops waiting for the authorization after about two minutes.

Click the button. LM Studio opens and adds the Aurora server. The button needs the LM Studio desktop app.

Add Aurora to LM Studio manually

In LM Studio, open the Program tab in the right sidebar.
Click Install, then Edit mcp.json.

Add Aurora to the mcpServers object:

{
  "mcpServers": {
    "aurora": {
      "url": "https://mcp.aurora.build/mcp"
    }
  }
}

Turn on the Aurora server. LM Studio opens your browser. Click Authorize within two minutes. Then load a model that supports tool calling and ask it to create an Aurora project.

Small local models often call tools incorrectly. Use a model with strong tool-calling support. For large changes, use the chat tool, so that Aurora’s own agent does the build.

Windsurf

Go to Windsurf Settings > MCP, or edit your MCP configuration file.

Add Aurora to your configuration:

{
  "mcpServers": {
    "aurora": {
      "serverUrl": "https://mcp.aurora.build/mcp"
    }
  }
}

When you first use an Aurora tool, Windsurf opens your browser. Click Authorize.

Available MCP tools

Once connected, Aurora exposes the following tools to your AI assistant:

Tool	Description
`get_started`	Get what Aurora can do, the workflow, and the project conventions
`create_project`	Create a new project from a description or from existing code
`list_projects`	List all your projects
`get_project`	Get details about a specific project
`get_file_tree`	View the project’s file structure
`read_file`	Read a file from a project
`read_files`	Read up to 20 files in one call
`apply_changes`	Write, edit, or delete several files in one commit
`chat`	Send a message to Aurora’s AI to make changes
`chat_status`	Get the result of a `chat` turn that is still running
`start_dev_server`	Start the project’s preview server
`restart_dev_server`	Restart the project’s preview server
`get_dev_logs`	Read the latest dev server output to find build errors
`deploy`	Deploy the project to production
`redeploy`	Rebuild and redeploy an existing deployment
`get_deployment_status`	Check the status of a deployment

Tips

Start in Claude, finish in Aurora. Design the first version in the chat. When you want it to work like a real app (saved data, user accounts, a public URL), ask your assistant to put it into Aurora.

Two ways to write code. Your AI assistant can write the code itself: it reads files with read_files, lands all changes for a task with one apply_changes call (one commit), and checks for build errors with get_dev_logs. Or it can hand the task to Aurora’s AI with chat. Only chat can add npm packages.

Long chat turns keep running. A turn can take several minutes. If it is not finished after about 40 seconds, chat returns a job ID, and your assistant calls chat_status to get the result. The turn continues on Aurora even if your tool disconnects.

Combine with other tools. MCP connections work alongside your existing workflow. Use Aurora to scaffold and deploy while using your local editor for fine-tuning.

Revoking access

To disconnect an MCP integration, go to Settings in Aurora and click Revoke next to the integration you want to remove.

Troubleshooting

OAuth prompt not appearing

Make sure your browser allows pop-ups from aurora.build. Some tools open the OAuth flow in a new tab or browser window.

Connection timeout

Verify you can reach https://mcp.aurora.build in your browser. If you’re behind a corporate firewall or VPN, you may need to allowlist the domain.

Tools not showing up

Restart your AI tool after adding the MCP server. Some tools require a reload to discover new MCP servers.

Permission denied errors

Your OAuth token may have expired. Remove and re-add the MCP server to re-authorize.

​What you can do via MCP

​Supported tools

Claude

Claude Code

Cursor

VS Code

LM Studio

Windsurf

​Before you connect

​Connect your tool

​Claude (claude.ai)

​Claude Code (CLI)

​Cursor

​VS Code

​LM Studio

​Windsurf

​Available MCP tools

​Tips

​Revoking access

​Troubleshooting

What you can do via MCP

Supported tools

Before you connect

Connect your tool

Claude (claude.ai)

Claude Code (CLI)

Cursor

VS Code

LM Studio

Windsurf

Available MCP tools

Tips

Revoking access

Troubleshooting