Async LLM Call with MCP Integration Example

This example demonstrates how to make asynchronous LLM calls with MCP (Multi-Component Platform) integration, including tool calling capabilities.

Prerequisites

• Python 3.8+
• aicore package installed
• MCP configuration file (JSON format)
• LLM provider API key (e.g., OpenAI, Deepseek, etc.)

Configuration

MCP Configuration (mcp_config.json)

{
  "mcpServers": {
    "search-server": {
      "transport_type": "ws",
      "url": "ws://localhost:8080"
    },
    "data-server": {
      "transport_type": "stdio",
      "command": "python",
      "args": ["data_server.py"]
    }
  }
}

LLM Configuration (config_example_mcp.yml)

llm:
  mcp_config_path: "./mcp_config.json"
  temperature: 0
  max_tokens: 8192
  provider: "deepseek"
  api_key: "your-api-key-here"
  model: "deepseek-chat"

Example Code

from aicore.config import Config
from aicore.llm import Llm
import os
import asyncio
import json

async def main():
    # Load configuration
    config = Config.from_yaml()  # Loads from config_example_mcp.yml by default
    
    # Print configuration details
    print("LLM Configuration:")
    print(json.dumps(config.llm.model_dump(), indent=4))
    print("\nMCP Configuration:")
    with open(config.llm.mcp_config_path) as f:
        print(json.dumps(json.load(f), indent=4))
    
    # Initialize LLM with MCP config
    llm = Llm.from_config(config.llm)
    
    # Make async completion with tool calling
    response = await llm.acomplete(
        "Search for latest AI news about Elon Musk",
        system_prompt="You are a helpful assistant with tool calling capabilities. "
                     "Use available tools when needed and provide comprehensive answers.",
        agent_id="mcp-agent",
        action_id="news-search"
    )
    
    print("\nFINAL RESPONSE:")
    print(response)

if __name__ == "__main__":
    asyncio.run(main())

Key Features Demonstrated

1. Asynchronous LLM Completion: Uses acomplete() for non-blocking API calls
2. MCP Server Integration: Automatically connects to configured MCP servers
3. Tool Calling Workflow:
   • Automatic tool discovery from connected MCP servers
   • Handling of tool call responses
   • Chaining multiple tool calls when needed
4. Observability Integration:
   • Tracks agent_id and action_id for monitoring
   • Records token usage and costs
5. Configuration Management:
   • Loads both LLM and MCP configurations
   • Validates required settings

Expected Output

The example will:

1. Print the loaded configuration
2. Connect to MCP servers and discover available tools
3. Execute the query using available tools
4. Print the final response which may include:
   • Direct LLM-generated content
   • Results from tool calls
   • Combined information from multiple sources

Troubleshooting

• Missing MCP Config: Ensure mcp_config_path is set in your LLM config
• Connection Issues: Verify MCP servers are running and accessible
• Tool Calling Errors: Check tool schemas match expected formats
• API Errors: Validate your LLM provider API key and model name

For more details see MCP Integration Documentation and Base Provider Documentation.