Design UML models and generate code with AI agents

Astah Pro MCP: Enabling AI-Powered UML Modeling

A local MCP server plugin for Astah Professional, a UML modeling tool, that enables AI agents to:

• Design systems and represent them as UML models.
• Explain UML models.
• Generate source code from UML models, and vice versa.
• Create UML diagrams from hand-drawn sketches.

This plugin experimentally implements a programmatic tool calling mode, inspired by the "Programmatic Tool Calling" idea described in this Anthropic article.

Demo

Prompts:
I. Design a simple library management system and draw a class diagram in Astah to illustrate its structure.
II. Based on the contents of the Astah model, draw a sequence diagram to illustrate the behavior of borrowing a book.
III. Based on the contents of the Astah model, draw an activity diagram to illustrate the behavior of returning a book.
IV. Based on the contents of the Astah model, draw a state machine diagram to illustrate the state transitions of a book.

Claude Code

Model: Opus 5 / Effort: Medium / Total time: approx. 13 min

I	II	III	IV

Codex CLI

Model: Astra / Effort: Medium / Total time: approx. 19 min

I	II	III	IV

Grok Build

Model: Grok 4.6 / Effort: Medium / Total time: approx. 41 min

I	II	III	IV

Requirements

• AI agent

  The AI agents below have been tested, but no guarantee is implied. From personal experience, Claude Code works well for UML modeling, followed by Codex CLI.

  • Claude Code
  • Codex CLI
  • Grok Build
  • Antigravity CLI
  • Cursor IDE
  • Antigravity IDE
  • Kiro IDE

• Astah Pro v12.0 or later

  Note: This MCP server only connects to AI agents running on the same machine as Astah Professional.
  According to Astah's terms of use, using Astah via an AI agent is permitted only if you hold a valid license and access it exclusively for your own use with your licensed Astah. Allowing a non-licensed third party to operate Astah via such an agent is strictly prohibited.
  For details, please refer to the FAQ (English / Japanese) or contact Change Vision (the developer of Astah) directly.

Supported Diagram Types

• Class Diagram
• Sequence Diagram
• Activity Diagram
• State Machine Diagram
• Use Case Diagram
• Mind Map
• Requirement Diagram
• Composite Structure Diagram (with some limitations)
• Communication Diagram (query-only)
• ER Diagram (query-only)

Installation

Install Astah Pro

Download from here and install.

Install the Astah Pro MCP plugin

Download the plugin JAR file (astah-pro-mcp-x.x.x.jar), drop it into Astah, and restart Astah (see here). If the mcp tab appears in the Extra View, the plugin is installed.

AI agent settings

This plugin experimentally implements a programmatic tool calling mode (port 8888), inspired by the "Programmatic Tool Calling" idea described in this Anthropic article. The direct tool calling mode also remains available (port 18888). Unless you have a specific reason to choose otherwise, specify port 8888.

claude mcp add --transport http --scope project astah-pro-mcp http://127.0.0.1:8888/mcp

claude mcp add --transport http --scope user astah-pro-mcp http://127.0.0.1:8888/mcp

[mcp_servers.astah-pro-mcp]
transport = "http"
url = "http://127.0.0.1:8888/mcp"

grok mcp add --scope project --transport http astah-pro-mcp http://127.0.0.1:8888/mcp

grok mcp add --scope user --transport http astah-pro-mcp http://127.0.0.1:8888/mcp

{
  "mcpServers": {
    "astah-pro-mcp": {
      "serverUrl": "http://127.0.0.1:8888/mcp"
    }
  }
}

{
  "mcpServers": {
    "astah-pro-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "http://127.0.0.1:8888/mcp",
        "--allow-http"
      ]
    }
  }
}

{
  "mcpServers": {
    "astah-pro-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "http://127.0.0.1:8888/mcp",
        "--allow-http"
      ]
    }
  }
}

{
  "mcpServers": {
    "astah-pro-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "http://127.0.0.1:8888/mcp",
        "--allow-http"
      ]
    }
  }
}

Changing the Port Numbers

By default, the plugin listens on 8888 (programmatic tool calling mode) and 18888 (direct tool calling mode). Each port can be overridden with an environment variable:

Remember to update your AI agent settings to the same port numbers.

How to Use

1. Start Astah Pro

   Some AI agents try to connect to the MCP server on startup, so start Astah Pro first.

2. Start AI agents

   Each time an AI agent establishes a new session with the Astah Pro MCP server, you will be asked to confirm. Review the details and click 'Connect'.

3. Send prompts to the AI agents

If you want to disable the Astah Pro MCP plugin in Astah, click [Plugin] > [Installed Plugins], select the Astah Pro MCP entry in the plugin list dialog, click Disable, and then restart Astah.

Build & Test

If you want to build and test locally:

1. Set up your Astah plugin development environment (see here).

2. Build:

   astah-build
   

3. Run tests (change astahPath to your Astah Pro installation path):

   Run all tests on Windows 11:

   astah-mvn test -DastahPath="C:\Program Files\astah-professional"
   

   Run specific tests on Windows 11:

   astah-mvn test -DastahPath="C:\Program Files\astah-professional" -Dtest="**/editor/*Test"
   

Considerations

Experimental Status

This project is currently experimental. The design and implementation may undergo breaking changes.

Confidentiality Risk

Astah project data and logs will be shared with the AI agent. For Astah projects that contain confidential information, either refrain from using this MCP server or use it only with appropriate safeguards (e.g., enabling opt-out settings for AI agents).

Backups Recommended

Because this MCP server edits model elements and diagrams, we recommend committing your Astah project to a Git repository or making copies before and during use so you can revert if necessary.

Feature Limitations

Some model or diagram information (e.g., certain properties) cannot be viewed or edited via the provided tool functions. Use the Astah GUI directly for those.

Costs

This MCP server prioritizes providing the information AI agents need and does not implement token-saving measures. We recommend using AI agents on a flat-rate plan rather than pay-as-you-go.

AI Limitations

Just as when working with source code, AI agents can make mistakes or misinterpret model elements and diagrams.

Got a feature request or found a bug?

Please open an issue. Because this project is in an experimental phase and may introduce breaking changes, we aren't accepting pull requests until the design and implementation stabilize. Thank you for your understanding.

Need support?

If you need private support, contact takaaki.teshima.dev [at] gmail.com (replace [at] with @). It could become a paid project; I may still be able to support you/it.