Pharos Ai Doc Genie

Built for Pharos Skill-to-Agent Dual Cascade Hackathon — Phase 1

A reusable, standardized Skill module that enables any AI Agent in the Pharos ecosystem to generate real Office documents (.pptx, .docx, .xlsx) and source code from natural language — powered by DashScope LLM API.

License: MIT Node.js MCP

---

🎯 Problem Statement

AI Agents in the Pharos economy need to produce tangible outputs — not just text responses. When an agent helps a user prepare a business proposal, it should deliver a real .docx file. When it analyzes data, it should produce an actual .xlsx spreadsheet. When it creates a presentation, it should output a .pptx that opens in PowerPoint.

Existing solutions either:

• Generate plain text/Markdown that requires manual formatting
• Depend on proprietary cloud APIs with unpredictable availability
• Lack standardized interfaces for agent-to-skill communication

Pharos AI Doc Genie fills this gap with a production-ready, standardized Skill that generates real Office files and code from natural language.

---

🧩 Skill Architecture

┌──────────────────────────────────────────────┐
│                AI Agent (Pharos)              │
│         (Any MCP-compatible Agent)            │
└─────────────────┬────────────────────────────┘
                  │ MCP Protocol (JSON-RPC 2.0)
                  │ stdio transport
┌─────────────────▼────────────────────────────┐
│          Pharos AI Doc Genie Skill            │
│                                               │
│  ┌──────────┐ ┌──────────┐ ┌──────────┐      │
│  │ generate │ │ generate │ │ generate │      │
│  │ _word    │ │  _ppt    │ │ _excel   │ ...  │
│  └────┬─────┘ └────┬─────┘ └────┬─────┘      │
│       │            │            │             │
│  ┌────▼────────────▼────────────▼──────┐      │
│  │        LLM (DashScope qwen)         │      │
│  │    Content Generation Layer         │      │
│  └────────────────┬────────────────────┘      │
│                   │                            │
│  ┌────────────────▼────────────────────┐      │
│  │     Python (python-pptx, etc.)      │      │
│  │     File Conversion Layer           │      │
│  │     Markdown → real .pptx/.docx     │      │
│  └────────────────┬────────────────────┘      │
└───────────────────┼───────────────────────────┘
                    │
         ┌──────────▼──────────┐
         │   Output Files      │
         │  .pptx  .docx       │
         │  .xlsx  .py/.js/... │
         └─────────────────────┘

Key design principles:

• Stateless: Each tool call is independent — no session state needed
• Idempotent: Same input produces consistent output structure
• Self-contained: Zero external service dependencies beyond the LLM API
• Standardized: MCP protocol ensures any compatible Agent can call it

---

🛠️ Tools (4 Skills)

Tool	Output	Use Case	Model
generate_ppt	.pptx presentation	Pitch decks, training, reports	qwen3.7-plus
generate_word	.docx document	Proposals, manuals, reports	qwen3.7-plus
generate_excel	.xlsx spreadsheet	Data tables, financials, inventory	qwen3.7-plus
generate_code	Source code (.py/.js/.go/...)	Rapid prototyping, boilerplate	qwen-long-latest

Tool Schema Examples

generate_ppt: Create a professional presentation

{
  "name": "generate_ppt",
  "arguments": {
    "topic": "AI in Enterprise: 2026 Trends",
    "requirements": "Executive summary for CTO audience, 12 slides, focus on ROI and adoption metrics",
    "slide_count": 12
  }
}

generate_excel: Generate structured data

{
  "name": "generate_excel",
  "arguments": {
    "description": "Q2 2026 sales data: Region, Product Category, Revenue, Units Sold, Growth%, Top Salesperson",
    "rows": 30
  }
}

---

🚀 Quick Start

Prerequisites

• Node.js >= 18
• Python 3.8+ with python-pptx, python-docx, openpyxl
• DashScope API Key (Alibaba BaiLian)

Install Python dependencies

pip install python-pptx python-docx openpyxl

Run the MCP Server

node src/mcp-server.js

The server listens on stdin/stdout using the MCP stdio transport. Configure your Agent's MCP client to launch this process.

Test with MCP Inspector

npx @modelcontextprotocol/inspector node src/mcp-server.js

Manual Test (JSON-RPC via pipe)

echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"test","version":"1.0"}}}' | node src/mcp-server.js

---

📁 Project Structure

pharos-ai-doc-genie/
├── src/
│   └── mcp-server.js       # MCP stdio server (self-contained)
├── convert.py               # Python: Markdown → .pptx/.docx/.xlsx
├── output/                  # Generated Office files
├── package.json             # Node.js project config
├── README.md                # This file
├── LICENSE                  # MIT License
└── .gitignore

---

🔌 Integration

With Claude Desktop

{
  "mcpServers": {
    "pharos-ai-doc-genie": {
      "command": "node",
      "args": ["/absolute/path/to/pharos-ai-doc-genie/src/mcp-server.js"]
    }
  }
}

With OpenAI Agents

The server uses standard MCP tool schemas that are directly compatible with OpenAI function calling format. Simply configure your Agent to launch the server as an MCP subprocess.

With Pharos Agents

Pharos Agents can call this Skill via the MCP protocol. Once the Skill is registered, Agents discover it through tools/list and call it through tools/call.

---

🧪 Testing

# List available tools
echo '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}' | node src/mcp-server.js

# Generate a Word document
echo '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"generate_word","arguments":{"topic":"Project Proposal: AI Chatbot","requirements":"A formal proposal for building an enterprise AI chatbot. Include: executive summary, technical approach, timeline, budget estimate.","length":"medium"}}}' | node src/mcp-server.js

# Generate code
echo '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"generate_code","arguments":{"requirement":"A Python async function that fetches data from a REST API with exponential backoff retry logic","language":"python","comments":"en"}}}' | node src/mcp-server.js

---

⚡ Performance

Tool	Avg. Response Time	Max Tokens	File Size
generate_word	~20-40s	16384	30-50 KB (.docx)
generate_ppt	~30-60s	16384	25-40 KB (.pptx)
generate_excel	~15-25s	16384	5-15 KB (.xlsx)
generate_code	~15-30s	16384	N/A (text)

---

🔒 Security

• No API key exposure: The DashScope API key is server-side only and never sent to Agents
• Input validation: All Agent inputs are validated before processing
• Output isolation: Generated files are written to a dedicated output directory
• No persistent state: Each tool call is isolated with no cross-call data leakage

---

🗺️ Roadmap

Phase 2 (Agent Arena)

• Deploy as a persistent Skill on Pharos chain
• On-chain billing per document generation
• NFT-based document ownership and verification
• Multi-agent collaborative document editing

Beyond

• PDF generation and manipulation
• Image-to-document conversion (OCR → formatted docx)
• Multi-language document templates
• Real-time collaborative editing via WebSocket

---

👤 Author

huimingchen081-beep (GitHub)

Built for the Pharos Skill-to-Agent Dual Cascade Hackathon — Phase 1 (Skill Hackathon).

---

📄 License

MIT License — see LICENSE for details.

---

🙏 Acknowledgments

• Pharos Network — for building the AI Agent economy infrastructure
• DashScope (Alibaba BaiLian) — for the LLM API powering content generation
• Model Context Protocol (Anthropic) — for the standardized agent-skill communication protocol
• python-pptx / python-docx / openpyxl — for Office file generation