Finance MCP Server

import asyncio import os import re from dotenv import load_dotenv from langchain_mcp_adapters.tools import load_mcp_tools from langchain_openai import ChatOpenAI from langgraph.prebuilt import create_react_agent from mcp import ClientSession, StdioServerParameters from mcp.client.stdio import stdio_client import warnings warnings.filterwarnings('ignore') # Environment variables load_dotenv() openai_api_key = os.environ["OPENAI_API_KEY"] model = ChatOpenAI(model="gpt-4o") # Parameters include how to run the server (using Python) and the program code file to execute server_params = StdioServerParameters( command="python", args=["mcp_server.py"], ) # The "main()" function is asynchronous, because calling the MCP tools is also asynchronous async def main(): async with stdio_client(server_params) as (read, write): async with ClientSession(read, write) as session: await session.initialize() tools = await load_mcp_tools(session) # Create a LangGraph ReAct (Reasoning and Acting) agent agent = create_react_agent(model, tools) agent_response = await agent.ainvoke({"messages":"What company uses the stock ticker META " \ "and how did this company's revenue develop over the last quarters and years?"}) # Define a regex pattern with: # - Literal match for ToolMessage(content=' # - A non-greedy capture group for everything up to the next ' pattern = r"ToolMessage\(content='(.*?)'" # Use re.findall with DOTALL so '.' matches newlines too matches = re.findall(pattern, str(agent_response), flags=re.DOTALL) # Display all extracted contents for i, content in enumerate(matches, 1): # Replace literal "\n" with actual newline characters content = content.replace("\\n", "\n") print(f"Match {i}:\n{content}\n") if __name__ == "__main__": asyncio.run(main())