A local MCP powered financial analyst agent using the DeepSeek-R1 model via Ollama, utilizing Github Copilot and Python. It analyzes user queries related to stocks and financial data, running entirely offline.

MCP Financial Analyst

MCP Financial Analyst is a local, multi-agent system built using the MCP. It enables natural language-driven financial analysis by leveraging CrewAI agents and running a local DeepSeek-R1 language model through Docker-hosted Ollama.

This project converts natural language queries into Python code that retrieves, analyzes, and visualizes stock market data, then securely executes the code and returns the results.

Features

• Natural language interface for financial data analysis
• Modular, agent-based architecture using CrewAI:
  • Query Parser Agent – interprets natural language queries
  • Code Writer Agent – generates Python code for analysis and plotting
  • Code Executor Agent – safely executes code in a sandbox
• Fully local LLM inference using DeepSeek-R1 via Docker-hosted Ollama
• Chart generation with pandas, yfinance, and matplotlib
• Runs locally using the MCP runtime

Prerequisites

• Python 3.10+
• Visual Studio Code
• Docker
• Ollama installed and running via Docker
• uv for running the MCP server (or use pip)

To start Ollama with DeepSeek-R1 in Docker:

docker run -it -p 11434:11434 --name give_name ollama/ollama
docker_terminal:
    ollama pull deepseek-r1
    exit

Ensure the Ollama server is up and listening on localhost:11434

Install Dependencies

Install Python dependencies using uv:

uv pip install -r requirements.txt

Running the MCP Server

You can run the server directly from VS Code terminal:

uv --directory /absolute/path/to/financial-analyst-deepseek run server.py

This will launch the MCP server that manages the CrewAI agents and handles user queries.

Example Query

Query:

Plot year-to-date performance for AAPL and MSFT

The system will:

1. Parse the query into structured instructions
2. Generate Python code to fetch stock data using yfinance
3. Plot the result using matplotlib
4. Exec