🤖 AI Orchestrator MCP Server (DeepSeek & Multi-LLM)

Production-ready Model Context Protocol (MCP) Server for delegating tasks from Claude Code to DeepSeek (deepseek-chat & deepseek-reasoner) and future LLM providers (Gemini, OpenAI, Qwen, Local LLMs).

🌟 Key Features

• Layered Architecture: Decoupled Tools, Router, Merger, Prompt Library, and Provider Abstraction.

• Provider Abstraction: Single unified AIProvider interface. Extendable to Gemini, OpenAI, Qwen without modifying tools or Claude Code integration.

• DeepSeek Integration: Full support for deepseek-chat and reasoning model deepseek-reasoner.

• Parallel Task Execution: Execute concurrent audits (e.g. Code + SQL + Security + Architecture) simultaneously.

• Result Merger Engine: Automatic deduplication, severity prioritization (CRITICAL -> LOW), and unified report generation.

• External Prompt Library: Markdown prompt templates stored in prompts/*.md.

• 20 Standalone MCP Tools:

  • Review & Analysis (14 Tools):

    • review_code

    • review_folder

    • review_project

    • review_sql

    • review_architecture

    • review_security

    • review_performance

    • write_tests

    • generate_seed

    • summarize

    • documentation

    • analyze_repository

    • explain_code

    • refactor_code

  • Scaffolding & Code Generation (6 Tools):

    • generate_code

    • generate_files

    • generate_sql

    • generate_tests

    • generate_documentation

    • generate_project

Related MCP server: mcp-agent-review

🚀 Setup & Configuration

1. Installation

npm install
npm run build

2. Configure Environment

Create .env file from .env.example:

DEEPSEEK_API_KEY=your_actual_deepseek_api_key
DEEPSEEK_BASE_URL=https://api.deepseek.com
DEEPSEEK_DEFAULT_CHAT_MODEL=deepseek-chat
DEEPSEEK_DEFAULT_REASONER_MODEL=deepseek-reasoner
DEEPSEEK_REASONING_EFFORT=max

DEFAULT_PROVIDER=deepseek
MAX_PARALLEL_TASKS=5
DEFAULT_TIMEOUT_MS=120000
MAX_RETRIES=3
LOG_LEVEL=info

3. Integration with Claude Code (claude_desktop_config.json or Claude Code config)

Add the following to your MCP settings:

{
  "mcpServers": {
    "ai-orchestrator": {
      "command": "node",
      "args": ["c:/Users/User/RiderProjects/DeepSeek-MCP/dist/index.js"],
      "env": {
        "DEEPSEEK_API_KEY": "your_deepseek_api_key_here"
      }
    }
  }
}

🛠️ Adding a New Provider (e.g. GeminiProvider)

1. Create src/providers/gemini/GeminiProvider.ts implementing AIProvider.

2. Register the new provider in src/providers/index.ts.

3. Set DEFAULT_PROVIDER=gemini or specify "provider": "gemini" when invoking MCP tools.

No changes to Claude Code or MCP Tool definitions are required!