aig-risk-profiler-mcp

A Model Context Protocol (MCP) server for generating Management Liability Coverage (MLC) risk analysis reports from Chinese company annual report PDFs.

Features

• analyze_company: Upload a company annual report PDF and start analysis
• check_task_status: Check the status and progress of an analysis task
• get_report: Retrieve the completed analysis as a PDF download URL or markdown content

Prerequisites

• Node.js >= 18.0.0
• The MLC Agent backend must be running (see backend setup instructions)

Installation

From npm

npm install -g aig-risk-profiler-mcp

Usage with Claude Desktop

Add the following to your Claude Desktop configuration file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json

Using npx (recommended)

{
  "mcpServers": {
    "aig-risk-profiler": {
      "command": "npx",
      "args": ["-y", "aig-risk-profiler-mcp"],
      "env": {
        "BACKEND_URL": "https://aig-risk-profiler.ragents.net"
      }
    }
  }
}

Using global install

{
  "mcpServers": {
    "aig-risk-profiler": {
      "command": "aig-risk-profiler-mcp",
      "env": {
        "BACKEND_URL": "http://localhost:8101"
      }
    }
  }
}

Using local build

{
  "mcpServers": {
    "aig-risk-profiler": {
      "command": "node",
      "args": ["/path/to/mcp-server/dist/index.js"],
      "env": {
        "BACKEND_URL": "http://localhost:8101"
      }
    }
  }
}

Environment Variables

| Variable | Description | Default | |----------|-------------|---------| | BACKEND_URL | URL of the MLC Agent backend API | http://localhost:8101 |

Tools

analyze_company

Upload a Chinese company annual report PDF and start MLC analysis.

Input:

• pdf_path (required): Local file path to the company annual report PDF

Output:

• task_id: ID to track the analysis task
• status: Current status of the task
• message: Human-readable message

Example:

User: 分析这个年报 /Users/me/Documents/紫光股份2023年报.pdf
AI: [calls analyze_company] 分析已开始，任务ID是 123

check_task_status

Check the status and progress of an analysis task. Use this to monitor if a task is still running, completed, or failed.

Input:

• task_id (required): The task ID returned by analyze_company

Output:

• task_id: The task ID
• status: "pending", "running", "completed", or "failed"
• progress: Human-readable progress message
• current_step: Name of the currently running step (if any)
• completed_steps: Number of completed steps
• total_steps: Total number of steps
• company_name: Name of the company being analyzed
• error: Error message if the task failed

Example:

User: 任务 123 完成了吗？
AI: [calls check_task_status] 正在运行：研究新闻和监管文件（7/10 步骤已完成）

get_report

Retrieve the MLC analysis report for a completed task.

Input:

• task_id (required): The task ID returned by analyze_company
• format (optional): Report format - "pdf" (default) or "markdown"

Output:

• status: "completed" or "not_completed"
• format: The report format ("pdf" or "markdown")
• download_url: URL to download the PDF (for pdf format)
• content: The markdown content (for markdown format)
• company_name: Name of the analyzed company
• message: Human-readable message

Example:

User: 给我报告
AI: [calls get_report] PDF报告已生成：http://localhost:8101/api/tasks/123/report/download

Typical Workflow

Analysis takes approximately 20 minutes. The workflow is designed to be asynchronous:

1. Start Analysis: Use analyze_company to upload a PDF and start the analysis
2. Check Progress: Use check_task_status periodically to monitor progress
3. Get Report: Once completed, use get_report to get the PDF download URL

User: 分析 /path/to/annual_report.pdf
AI: [calls analyze_company] 分析已开始，任务ID是 123，预计需要20分钟

User: 好了吗？
AI: [calls check_task_status] 正在运行：研究新闻和监管文件（7/10 步骤已完成）

User: 现在呢？
AI: [calls check_task_status] 已完成！

User: 给我报告
AI: [calls get_report] PDF报告下载链接：http://localhost:8101/api/tasks/123/report/download

Development

# Install dependencies
pnpm install

# Run in development mode
pnpm dev

# Type check
pnpm typecheck

# Build for production
pnpm build

# Run production build
pnpm start

PDF Reports

PDF reports are generated by the backend and include:

• Company overview
• Financial analysis
• Governance structure
• Shareholder information
• News and regulatory findings
• Risk assessment
• Works cited / data sources

The PDF is professionally styled with:

• Clear typography and hierarchy
• Table formatting
• Risk indicators
• Print-optimized layout

Troubleshooting

Backend connection refused

Make sure the MLC Agent backend is running:

cd backend
docker compose up -d
# or
uv run uvicorn mlc_agent.main:app --port 8101

Task timeout

Analysis can take several minutes for large PDFs. Use check_task_status to monitor progress. If a task takes too long, check the backend logs.

Changelog

v1.1.0

• Added check_task_status tool for monitoring task progress
• Moved PDF generation from MCP to backend
• Simplified get_report tool - now returns download URL instead of generating PDF locally
• Removed wait_for_completion parameter from get_report
• Removed md-to-pdf dependency
• Removed OUTPUT_DIR environment variable

v1.0.0

• Initial release
• analyze_company tool for uploading PDFs
• get_report tool with local PDF generation

License

MIT