MCP Securities Analysis

A Python-based flow for securities analysis using the Model Context Protocol (MCP). The repository bundles data-collection, parsing, analytics and visualisation tools behind a single FastMCP server so that they can be consumed locally or remotely by any MCP-aware client (e.g. Claude Desktop, LangChain, OpenAI-Function calling, etc.).

Example deep research report for Tesla.

This was generated semi-autonomously by the following steps:

• connect MCP tools to Claude Desktop, including web search, Perplexity, Wikipedia, in addition to the market data tools in server.py for fundamental, technical analysis, and news search.
• prompt Claude Desktop to query Perplexity, Wikipedia, and the 10-K to write a profile of Tesla
• prompt Claude Desktop to query each tool for info on Tesla
• finally, enable deep research and prompt Claude Desktop to write a deep report in 8 sections with details on what each section should cover, using the information retrieved from the tools.

While it's not a fully autonomous agent and at an early POC level, it shows clear path toward a fully autonomous agent. Create an MCP client that goes through the steps above and generates a deep report on Tesla in a structured format with graphs and tables. And then create an even more advanced multi-agent workflow with a set of parallel agents for each section, and a critic-optimizer workflow, and a final report generator.

---

Features

• FastMCP server – exposes a few MCP tools to get market data, news, charts, SEC filings, fundamental, technical data, research from public web sites, subscription services, and REST APIs.
• Market data – real-time and historical OHLCV data via yfinance & OpenBB.
• Fundamental data – automatic downloading of SEC filings (sec_downloader) and rich XBRL/HTML parsing through sec_parser.
• News & Social sentiment – headlines with newsapi-python plus Reddit scraping utilities.
• Technical analysis – hundreds of indicators with pandas_ta & TA-Lib.
• Interactive plots – high-quality Plotly charts exported server-side (static PNG or interactive HTML).
• Async-first design – built on asyncio, aiohttp, httpx & Playwright for maximum throughput.

• this section AI-generated so beware of hype. New project, would like to share and get comments, not extensively tested. Use it as a starting point, at your own risk.

---

Quick Start

# 1. Clone and enter the project
$ git clone https://github.com/<your-org>/MCP.git
$ cd MCP

# 2. Create & activate a virtualenv (recommended)
$ python -m venv .venv
$ source .venv/bin/activate

# 3. Install python dependencies
$ pip install -r requirements.txt

# 4. Install Playwright browsers (once)
$ playwright install

# 5. Copy environment template & add your keys
$ cp dotenv.txt .env  # then edit as needed

# 6. Launch and test the server
$ LOGLEVEL=DEBUG mcp dev server.py
# click to the link in the terminal to open the test page, connect, view tools, and then test them individually

# 7. Use the server in your MCP client of choice. For Claude Desktop, install the provided claude_desktop_config.json file for your platform (macOS, Windows).

• MCP / Claude Desktop Quickstart

• Awesome MCP Servers

---

Project Structure

MCP/
├── claude_desktop_config.json # Configuration for Claude Desktop
├── dotenv.txt             # Secrets / Environment variables
├── README.md              # This file
├── server.py              # FastMCP server, launched by mcp dev or Claude desktop or other MCP client
├── requirements.txt       # Python dependencies
├── sources.yaml           # Data-source configuration used by server
├── Market Data.ipynb      # Jupyter notebook to fetch market data
├── TearSheet.ipynb        # Jupyter notebook to do basic analysis