|
|
|
|
|
from contextlib import asynccontextmanager |
|
|
|
|
|
from langchain_mcp_adapters.client import MultiServerMCPClient |
|
|
from langchain_google_genai import ChatGoogleGenerativeAI |
|
|
from langgraph.checkpoint.memory import InMemorySaver |
|
|
from langgraph.prebuilt import create_react_agent |
|
|
|
|
|
import os |
|
|
from dotenv import load_dotenv |
|
|
|
|
|
load_dotenv() |
|
|
|
|
|
model = ChatGoogleGenerativeAI(model="gemini-2.0-flash", api_key=os.getenv("GOOGLE_API_KEY")) |
|
|
|
|
|
checkpointer = InMemorySaver() |
|
|
|
|
|
prompt = """ |
|
|
## Role |
|
|
You are an AI Developer specializing in AI frameworks, capable of creating Agents, chatbots, and Retrieval-Augmented Generation (RAG) systems. |
|
|
|
|
|
## Task |
|
|
Answer user questions and Write complete, production-ready code for AI Agents, chatbots, and RAG systems using framework documentation. |
|
|
|
|
|
## Instructions |
|
|
Follow these steps when fulfilling user request: |
|
|
|
|
|
1. Use the `list_doc_sources` tool to retrieve available documentations links. |
|
|
2. Call `fetch_docs` tool to analyze the content of the documentaion. |
|
|
3. Reflect on the URLs in the documentaions content and select the most pertinent URLs based on the content. |
|
|
4. Call `fetch_docs` tool on the selected URLs. |
|
|
5. Provide a clear and complete response to the user. |
|
|
6. If the current information is insufficient, fetch more URLs until the request is fulfilled. |
|
|
|
|
|
## Constraints |
|
|
Ensure your answers are correct, the code is accurate, production-ready, and fully leverages the documentation. |
|
|
|
|
|
""" |
|
|
@asynccontextmanager |
|
|
async def make_graph(): |
|
|
async with MultiServerMCPClient( |
|
|
{ |
|
|
"docs_mcp": { |
|
|
|
|
|
"url": "http://localhost:8082/sse", |
|
|
"transport": "sse", |
|
|
} |
|
|
} |
|
|
) as client: |
|
|
agent = create_react_agent( |
|
|
model, |
|
|
client.get_tools(), |
|
|
prompt=prompt, |
|
|
checkpointer=checkpointer) |
|
|
|
|
|
yield agent |
