finance-tools-mcp: A Financial Analysis MCP Server

https://github.com/VoxLink-org/finance-tools-mcp

Overview

The finance-tools-mcp is a Model Context Protocol (MCP) server designed to provide comprehensive financial insights and analysis capabilities to Large Language Models (LLMs). Modified from investor-agent, it integrates with various data sources and analytical libraries to offer a suite of tools for detailed financial research and analysis.

Tools Offered

The server exposes a variety of tools via the MCP, allowing connected clients (like LLMs) to access specific financial data and perform analyses:

Ticker Data Tools:
- get_ticker_data: Provides a comprehensive report for a given ticker, including overview, news, key metrics, performance, dates, analyst recommendations, and upgrades/downgrades.
- get_options: Retrieves options data with the highest open interest for a ticker, with filtering options for date range, strike price, and option type (Calls/Puts).
- get_price_history: Fetches historical price data digest for a specified period, including OHLCV samples, Technical Indicators, Risk Metrics, and other quantitative analysis.
- get_financial_statements: Accesses financial statements (income, balance, cash flow) for a ticker, available quarterly or annually.
- get_institutional_holders: Lists major institutional and mutual fund holders for a ticker.
- get_earnings_history: Provides earnings history with estimates and surprises for a ticker.
- get_insider_trades: Retrieves recent insider trading activity for a ticker.
- get_ticker_news_tool: Fetches the latest Yahoo Finance news for a specific ticker.
Fear & Greed Index Tools:
- get_current_fng_tool: Gets the current CNN Fear & Greed Index score and rating.
- get_historical_fng_tool: Retrieves historical CNN Fear & Greed Index data for a specified number of days.
- analyze_fng_trend: Analyzes trends in the CNN Fear & Greed Index over a specified period.
Calculation Tools:
- calculate: Evaluates mathematical expressions using Python's math syntax and NumPy.
Macro Data Tools:
- get_current_time: Provides the current time.
- get_fred_series: Retrieves data for a specific FRED series ID.
- search_fred_series: Searches for popular FRED series by keyword.
- cnbc_news_feed: Fetches the latest breaking world news from CNBC, BBC, and SCMP.

Time Series Data Processing and Optimization

The server utilizes yfinance to retrieve historical price data (OHLCV - Open, High, Low, Close, Volume) for tickers. This raw data undergoes significant processing and analysis to provide valuable insights, particularly optimized for consumption by LLMs.

Key aspects of the time series data handling include:

Comprehensive Analysis: The data is analyzed using libraries like ta-lib-python to calculate a wide range of technical indicators. Additionally, custom functions compute basic statistics, risk metrics, recognize common chart patterns, and calculate Fibonacci retracement levels.
Structured Digest: The results of this analysis are compiled into a structured digest format (generate_time_series_digest_for_LLM) that is easy for LLMs to parse and understand, including sections for statistics, summary, technical indicators, risk metrics, patterns, Fibonacci levels, and a data sample.
Smart Sampling for LLMs: To provide LLMs with a representative view of historical data without overwhelming the context window, a "smart sampling" strategy is employed (get_latest_data_sample). This method samples the data with varying resolutions:
- High Resolution: Recent data points are included daily.
- Medium Resolution: Intermediate data points are sampled weekly.
- Low Resolution: Older data points are sampled monthly. This hybrid approach ensures that the LLM receives detailed information about recent price movements while still having context about longer-term trends, all within a manageable number of data points.

This optimized processing and presentation of time series data allows LLMs to quickly grasp key trends, indicators, and patterns, facilitating more informed financial analysis.

Sample Report

alt text

Prerequisites

Python: 3.10 or higher
Package Manager: uv

Installation

First, install uv if you haven't already:

curl -LsSf https://astral.sh/uv/install.sh | sh

Then, you can run the finance-tools-mcp MCP server using uvx:

uvx finance-tools-mcp

If you want to use your own FRED API key, you can set it as an environment variable:

FRED_API_KEY=YOUR_API_KEY uvx finance-tools-mcp

You can also run the server using Server-Sent Events (SSE) transport:

uvx finance-tools-mcp --transport sse

Or with the FRED API key and SSE transport:

FRED_API_KEY=YOUR_API_KEY uvx finance-tools-mcp --transport sse

Usage with MCP Clients

To integrate finance-tools-mcp with an MCP client (for example, Claude Desktop), add the following configuration to your claude_desktop_config.json:

{
  "mcpServers": {
    "investor": {
        "command": "path/to/uvx/command/uvx",
        "args": ["finance-tools-mcp"],
    }
  }
}

Debugging

You can leverage the MCP inspector to debug the server:

npx @modelcontextprotocol/inspector uvx finance-tools-mcp

npx @modelcontextprotocol/inspector uv --directory  ./ run finance-tools-mcp

For log monitoring, check the following directories:

macOS: ~/Library/Logs/Claude/mcp*.log
Windows: %APPDATA%\Claude\logs\mcp*.log

Development

For local development and testing:

Use the MCP inspector as described in the Debugging section.
Test using Claude Desktop with this configuration:

{
  "mcpServers": {
    "investor": {
      "command": "path/to/uv/command/uv",
      "args": ["--directory", "path/to/finance-tools-mcp", "run", "finance-tools-mcp"],
    }
  }
}

License

This MCP server is licensed under the MIT License. See the LICENSE file for details.

Todo

Add supporting levels and resistance levels for stocks
Add Fibonacci retracement levels for stocks
Add moving average confluence levels for stocks
Add option model for prediction
Add predictive model by using finance sheets and other features

Data Sources

fintel.com
investing.com
yahoo.com
fred.stlouisfed.org
cnn cnbc and reddit