Conversational Agent
I'd like to create a simple conversational agent on GCP that leverages the mcp server that was created
To build a conversational AI agent that orchestrates tools from your sec-agent-mcp Cloud Run deployment, the standard production design uses a LangGraph ReAct Agent powered by gemini-2.5-flash via the langchain-mcp-adapters library.
This setup allows Gemini to dynamically inspect your remote MCP server's JSON-RPC tool schemas (lookup_cik, search_securities, search_filing_vectors) over SSE and invoke them in a conversational loop.
Install Dependencies in Cloud Shell
In your local environment or Cloud Shell, install the MCP bridge and Gemini LangChain integration:
requirements.txt
langchain-mcp-adapters>=0.1.0
langgraph>=0.2.0
langchain-google-genai>=2.0.0
google-genai>=0.3.0
python-dotenv>=1.0.0
pip install -r requirements.txt
Create agent.py
Create a script named agent.py in your project folder. It connects to [https://sec-agent-mcp-apkiqkbzyq-ue.a.run.app/sse](https://sec-agent-mcp-apkiqkbzyq-ue.a.run.app/sse), fetches your tools dynamically, and initializes the interactive agent loop:
import os
import warnings
import asyncio
# Suppress LangChain / LangGraph deprecation & serialization warnings
warnings.filterwarnings("ignore", category=UserWarning)
warnings.filterwarnings("ignore", category=DeprecationWarning)
from langchain_mcp_adapters.client import MultiServerMCPClient
from langchain.agents import create_agent
from langchain_google_genai import ChatGoogleGenerativeAI
CLOUD_RUN_SSE_URL = "https://sec-agent-mcp-apkiqkbzyq-ue.a.run.app/sse"
async def run_conversational_agent():
api_key = os.environ.get("GEMINI_API_KEY")
if not api_key:
raise ValueError("Please set your GEMINI_API_KEY environment variable.")
print(f"Connecting agent to MCP Server at {CLOUD_RUN_SSE_URL}...")
# Initialize connection with timeouts for Cloud Run latency
client = MultiServerMCPClient({
"sec_tools": {
"transport": "sse",
"url": CLOUD_RUN_SSE_URL,
"timeout": 30.0,
"sse_read_timeout": 300.0
}
})
# Automatically load remote tools exposed by FastMCP
tools = await client.get_tools()
print(f"Discovered {len(tools)} remote MCP tools: {[t.name for t in tools]}")
# Initialize Gemini 2.5 Flash
llm = ChatGoogleGenerativeAI(
model="gemini-2.5-flash",
google_api_key=api_key,
temperature=0.2
)
system_prompt = (
"You are an expert SEC EDGAR Financial Research Assistant. "
"Use the available MCP tools to lookup company CIKs, search securities master records, "
"and retrieve vector search filing chunks before answering quantitative finance questions."
)
# Replaced create_react_agent with create_agent and system_prompt
agent = create_agent(llm, tools, system_prompt=system_prompt)
print("\n--- SEC EDGAR Conversational Agent Ready ---")
print("Type your financial query below (or type 'exit' to quit):\n")
conversation_history = []
while True:
try:
user_input = input("User > ")
except (KeyboardInterrupt, EOFError):
break
if user_input.strip().lower() in ["exit", "quit"]:
break
if not user_input.strip():
continue
conversation_history.append(("user", user_input))
response = await agent.ainvoke({"messages": conversation_history})
# Process message response cleanly
last_msg = response["messages"][-1]
final_content = last_msg.content if isinstance(last_msg.content, str) else str(last_msg.content)
conversation_history.append(("assistant", final_content))
print(f"\nAgent > {final_content}\n")
if __name__ == "__main__":
asyncio.run(run_conversational_agent())
Run the Conversational Agent
Execute agent.py in your terminal:
$ python agent.py
/home/john_iacovacci1/.local/lib/python3.12/site-packages/langgraph/checkpoint/serde/encrypted.py:5: LangChainPendingDeprecationWarning: The default value of `allowed_objects` will change in a future version. Pass an explicit value (e.g., allowed_objects='messages' or allowed_objects='core') to suppress this warning.
from langgraph.checkpoint.serde.jsonplus import JsonPlusSerializer
Connecting agent to MCP Server at https://sec-agent-mcp-apkiqkbzyq-ue.a.run.app/sse...
Discovered 3 remote MCP tools: ['lookup_cik', 'search_securities', 'search_filing_vectors']
--- SEC EDGAR Conversational Agent Ready ---
Type your financial query below (or type 'exit' to quit):
User > Find the CIK for Apple and check if Microsoft is listed in our securities table.
Agent > The CIK for Apple Inc. is 0000320193. Yes, Microsoft is listed in the securities table with the ticker MSFT, FIGI BBG000BPH459, and is a Common Stock in the Equity market sector.
User >
No comments:
Post a Comment