Bring Your Own Agentic Framework with Llama Stack

pip install langgraph==0.6.7 langchain-openai==0.3.32 langchain-core==0.3.75

cat << 'EOF' > langgraph_llama_stack.py
import os
from llama_stack_client import LlamaStackClient
from langgraph.graph import StateGraph, END, START
from langchain_openai import ChatOpenAI
from langchain_core.messages import HumanMessage, AIMessage
from typing import Annotated
from typing_extensions import TypedDict
from langgraph.graph.message import add_messages

# Environment variables
INFERENCE_MODEL = os.getenv("INFERENCE_MODEL", "meta-llama/Llama-3.2-3B-Instruct")
INFERENCE_SERVER_OPENAI = os.getenv("LLAMA_STACK_ENDPOINT_OPENAI", "http://localhost:8321/v1/openai/v1")
API_KEY = os.getenv("OPENAI_API_KEY", "not-applicable")

print("📋 LangGraph + Llama Stack Integration")
print(f"   LLAMA_STACK_URL: {INFERENCE_SERVER_OPENAI}")
print(f"   INFERENCE_MODEL: {INFERENCE_MODEL}")

llm = ChatOpenAI(
    model=INFERENCE_MODEL,
    openai_api_key=API_KEY,
    openai_api_base=INFERENCE_SERVER_OPENAI,
    use_responses_api=True,
)

# Test connectivity
print("\n🧪 Testing basic connectivity:")
response = llm.invoke("Hello")
print(f"✅ Connection successful")

# Bind MCP weather tools
print("\n🛠️ Setting up MCP weather tools...")
llm_with_tools = llm.bind_tools([
    {
        "type": "mcp",
        "server_label": "weather",
        "server_url": "http://host.containers.internal:3001/sse",
        "require_approval": "never",
    },
])
print("✅ MCP tools configured")

# Define LangGraph State and Agent
class State(TypedDict):
    messages: Annotated[list, add_messages]

def chatbot(state: State):
    message = llm_with_tools.invoke(state["messages"])
    return {"messages": [message]}

# Build LangGraph StateGraph
print("\n🏗️ Building LangGraph agent...")
graph_builder = StateGraph(State)
graph_builder.add_node("chatbot", chatbot)
graph_builder.add_edge(START, "chatbot")
graph_builder.add_edge("chatbot", END)

graph = graph_builder.compile()
print("✅ LangGraph agent ready")

# Test the integration
print("\n" + "="*50)
print("🚀 Testing LangGraph Agent with MCP Tools")
print("="*50)

response = graph.invoke({
    "messages": [{"role": "user", "content": "What is the weather in Seattle?"}]
})

print("Weather Response:")
for message in response['messages']:
    if hasattr(message, 'content'):
        if isinstance(message.content, list):
            for content_block in message.content:
                if content_block.get('type') == 'text':
                    print(content_block.get('text', ''))
        elif isinstance(message.content, str):
            print(message.content)
    else:
        message.pretty_print()
EOF

python langgraph_llama_stack.py