​
Overview

The Intent Classifier pattern provides sophisticated natural language understanding to categorize user requests. It is a close sibling of the router workflow.

​
Quick Example

from mcp_agent.app import MCPApp
from mcp_agent.workflows.intent_classifier.intent_classifier_base import Intent
from mcp_agent.workflows.intent_classifier.intent_classifier_llm_openai import OpenAILLMIntentClassifier
from mcp_agent.workflows.intent_classifier.intent_classifier_embedding_openai import OpenAIEmbeddingIntentClassifier

app = MCPApp(name="intent_example")

async with app.run() as intent_app:
    context = intent_app.context

    embedding_intent_classifier = OpenAIEmbeddingIntentClassifier(
        intents=[
            Intent(
                name="greeting",
                description="A friendly greeting",
                examples=["Hello", "Hi there", "Good morning"],
            ),
            Intent(
                name="farewell",
                description="A friendly farewell",
                examples=["Goodbye", "See you later", "Take care"],
            ),
        ],
        context=context,
    )

    results = await embedding_intent_classifier.classify(
        request="Hello, how are you?",
        top_k=1,
    )

    print(f"Embedding-based Intent classification results: {results}")

    llm_intent_classifier = OpenAILLMIntentClassifier(
        intents=[
            Intent(
                name="greeting",
                description="A friendly greeting",
                examples=["Hello", "Hi there", "Good morning"],
            ),
            Intent(
                name="farewell",
                description="A friendly farewell",
                examples=["Goodbye", "See you later", "Take care"],
            ),
        ],
        context=context,
    )

    results = await llm_intent_classifier.classify(
        request="Hello, how are you?",
        top_k=1,
    )

    print(f"LLM-based Intent classification results: {results}")

​
Key Features

  • Multiple Classification Methods: Choose between embedding-based or LLM-based classification
  • Custom Intent Definitions: Define intents with descriptions and example phrases
  • Top-K Results: Return multiple potential intents ranked by confidence
  • Flexible Integration: Easy integration with existing chat and routing systems

​
Use Cases

  • Customer Support Routing: Automatically route tickets to appropriate teams
  • Chatbot Intelligence: Provide contextually relevant responses
  • Content Personalization: Customize content based on user intent
  • Analytics and Insights: Track user intent patterns and trends

Full Implementation

See the complete intent classifier with hierarchical categories and confidence scoring.