Article written and saved. Here's a summary:

**File:** `articles/astah-pro-mcp-uml-ai-agent.html` (891 words)

**Key facts from the source:**

- Created by Takaaki Teshima, released under MIT-0

- Local MCP server plugin for Astah Professional v12.0+

- Two modes: Programmatic tool calling (port 8888, 83 tools) and Direct (port 18888, 398 tools)

- Supports class, sequence, activity, state machine, use case, mind map, requirement, composite structure diagrams; communication and ER are query-only

- Tested with Claude Code, Codex CLI, Grok Build, Antigravity CLI, Cursor IDE, Antigravity IDE, Kiro IDE

- Demo: library management system across 4 chained prompts; Claude Code/Opus 5 finished in ~13 min, Codex CLI/Astra in ~19 min, Grok Build/Grok 4.6 in ~41 min

- Shares project data with the AI agent; no token-saving measures; recommends flat-rate plans

- 17 stars, 5 forks, 44 commits, experimental status

**Structure:**

1. **Opening** - What the plugin does and why it matters

2. **How It Works** - The two connection modes, why 83 tools beats 398, HTTP vs stdio

3. **The Demo** - Three agents running the same 4-prompt library management sequence, timing comparison

4. **What the Agent Can and Cannot Do** - Feature coverage, limits, token volume, confidentiality

5. **Supported Agents and Licensing** - Tested agents, Astah licensing terms, MIT-0

6. **Where It Stands** - Experimental status, project size, the programmatic tool calling question