Skip to main content

Using Axure RP with AI tools via MCP

info

The MCP server in Axure RP is currently experimental. Features and behavior may change as we continue to develop it.

Axure RP includes a built-in MCP server that gives AI tools read-only access to the files you have open in Axure RP. Once set up, you can ask your AI tool about the pages, widgets, interactions, and notes in your current file, and it will read that content directly from Axure RP, using your design as context for code generation, analysis, documentation, and more.

MCP stands for Model Context Protocol, an open standard for giving AI agents and tools access to external data sources. Because the MCP server in Axure RP runs locally on your computer, only AI applications running on the same computer can connect to it. AI tools that run in the browser, like ChatGPT or Claude on the web, can't connect: they run on the provider's servers, which have no access to your desktop. This page includes setup instructions for the Claude Desktop and ChatGPT desktop apps, as well as Claude Code, VS Code, and Cursor.

info

The MCP server in Axure RP is separate from the Axure Cloud MCP server, which gives AI tools access to whiteboards and documents hosted on Axure Cloud. Use the MCP server in Axure RP when you want AI tools to read the .rp files you're working on locally.

How it works​

When the MCP server is enabled, Axure RP hosts it on your computer at the following address:

http://127.0.0.1:41414/mcp

AI tools connected to this address can read the contents of the .rp files currently open in Axure RP: the sitemap, the widgets on each page, interactions, page and widget notes, and rendered images of pages and widgets. When you close a file or exit Axure RP, its content is no longer available to the AI tool.

The server is only reachable from your own computer. It is not exposed to your network or the internet.

info

All access through MCP is currently read-only: AI tools can inspect your files but cannot modify them.

Requirements​

  • Axure RP 11 (build 11.0.0.4147 or later) on Windows or macOS
  • An MCP-compatible AI application installed on the same computer
  • For the Claude Desktop setup below: Node.js, which provides the npx command

Enabling the MCP server​

  1. In Axure RP, go to File → MCP Settings... to open the MCP tab of the preferences dialog. You can also open the dialog via File → Preferences (F9) on Windows or Axure RP → Settings on Mac and then select the MCP tab.

  2. Check Enable MCP server.

The server starts right away, and you don't need to restart Axure RP. To stop the server, uncheck Enable MCP server.

warning

The MCP server does not require authentication. It can only be reached from your own computer, but while it is enabled, any application running on your computer can read the contents of your open files. If you work with sensitive designs, enable the server only while you're using it.

Connecting your AI tool​

The steps vary depending on which tool you're using. In all cases, Axure RP must be running with the MCP server enabled for the tool to connect.

Claude Desktop​

Claude Desktop connects to local MCP servers through its configuration file, using the mcp-remote bridge to reach the server's local address.

  1. In Claude Desktop, go to Settings → Developer and click Edit Config. This opens the folder containing the claude_desktop_config.json configuration file.

  2. Open claude_desktop_config.json in a text editor, add the following, and save the file:

{
  "mcpServers": {
    "axure-rp": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "http://127.0.0.1:41414/mcp",
        "--allow-http"
      ]
    }
  }
}

  1. Quit and reopen Claude Desktop.

  2. To verify the connection, click the Search and tools icon under the message box. You should see axure-rp in the list of available tools.

info

This setup requires Node.js. Claude Desktop's custom connectors, added via Settings → Connectors, only work with servers on the public internet, so they can't connect to the local server in Axure RP. The configuration file method above works with local servers.

ChatGPT desktop app​

The ChatGPT desktop app can connect to the MCP server directly. These steps only apply to the desktop app: ChatGPT in the browser cannot reach servers running on your computer, so download the desktop app if you don't have it yet.

  1. In the ChatGPT desktop app, go to Settings → MCP servers and click Add server.

  2. Enter a name, such as Axure RP.

  3. For the server type, choose Streamable HTTP.

  4. For the URL, enter http://127.0.0.1:41414/mcp.

  5. Save the server and restart the app if prompted.

info

MCP server support in ChatGPT may require developer mode or a specific plan. See OpenAI's MCP documentation for the current requirements.

Claude Code​

Run the following command in your terminal:

claude mcp add --transport http axure-rp http://127.0.0.1:41414/mcp

You can verify the server was added by running claude mcp list.

VS Code​

Run the MCP: Open User Configuration command in VS Code to open your mcp.json file, then add the Axure RP server to the servers object:

{
  "servers": {
    "axure-rp": {
      "type": "http",
      "url": "http://127.0.0.1:41414/mcp"
    }
  }
}

To make the server available in a single workspace instead, add the same entry to a .vscode/mcp.json file in that workspace.

Cursor​

In Cursor, go to Settings → Cursor Settings → MCP and add a new server, or add the Axure RP server to your ~/.cursor/mcp.json file directly:

{
  "mcpServers": {
    "axure-rp": {
      "url": "http://127.0.0.1:41414/mcp"
    }
  }
}

Other MCP-compatible tools​

Any AI tool that runs on your computer and supports MCP over streamable HTTP can connect to Axure RP using the following server URL:

http://127.0.0.1:41414/mcp

Refer to your tool's documentation for how to add an MCP server.

Using Axure RP with AI tools​

Once your AI tool is connected, you can ask it about your open files in plain language. The server provides tools for reading:

  • The list of files currently open in Axure RP
  • The sitemap and the component list of a file
  • The widgets on a page or component, including their positions, text, and styling
  • The interactions on a page, component, or individual widget
  • Page notes, widget notes, and note fields
  • Rendered images of pages, dynamic panel states, and individual widgets
  • The widgets you currently have selected on the canvas

You don't need to tell the AI tool which file to look at. When you refer to "my current file" or "this page," the server resolves that to the file you most recently worked in, and each response includes the file name so you can confirm it read the right one.

Example prompts​

Here are some prompts to get you started.

Generate components
Look at the page I currently have open in Axure RP.

Return production-ready React components that match the layout. Structure your response as:
1. Component list (name and one-line description of each)
2. Code for each component (semantic HTML, TypeScript props)
3. Interactive states and variants (note any you've inferred)

If a detail is unclear, state your assumption in [brackets] and continue.
Summarize interactions
Look at the page I currently have open in Axure RP.

Return a markdown document describing every interaction on the page. For each, include:
- The widget it's attached to
- The triggering event
- Conditions, if any
- The resulting actions, in order

Flag any widgets that look interactive (buttons, links, form fields) but have no interactions attached.
Audit notes and annotations
Look at the page I currently have open in Axure RP.

Review the page notes and widget notes. Return a markdown table with these columns: Widget | Current note | Issue | Suggested rewrite

Then list any widgets that likely need a note but don't have one (form fields without validation notes, buttons without behavior descriptions, and so on).
Write a functional spec
Look at the sitemap of the file I currently have open in Axure RP, then read each page.

Return a markdown functional specification with one section per page. For each page, include:
1. Purpose (1 sentence)
2. Key UI elements (bullet list)
3. Behaviors and interactions
4. Open questions (anything ambiguous in the design)

Troubleshooting​

Check that the server is running. With Axure RP open and the MCP server enabled, visit http://127.0.0.1:41414/health in a browser. If the server is running, the page displays ok.

Check the server's address. If another application is using port 41414, Axure RP uses the next available port instead. The server's actual address is recorded in a mcp-server.json file while the server is running:

  • Windows: %LOCALAPPDATA%\Axure\mcp-server.json
  • Mac: ~/.local/share/Axure/mcp-server.json

If the url in that file differs from http://127.0.0.1:41414/mcp, use that URL in your AI tool's configuration instead.

Screenshots aren't visible in the conversation. Some AI chat clients don't display the images returned by MCP tools directly in the conversation, even though the AI can see them. Look for a collapsed tool call entry in the transcript and expand it to view the image.