基于 Python 的证券分析流程，使用 模型上下文协议（MCP）。该仓库将市场数据、解析、分析和可视化工具捆绑在 FastMCP 服务器后面，使任何支持 MCP 的客户端（如 Claude Desktop、LangChain）都能本地或远程调用。

特斯拉的深度研究报告示例。

该报告通过以下步骤半自主生成：

• 将MCP工具连接到Claude Desktop，包括网页搜索、Perplexity、维基百科。（详情见claude_desktop_config.json）。你可以把维基百科工具看作是你邀请的朋友，他懂得如何搜索和导航维基百科，并将文章、章节、摘要带入聊天环境，以便与用户和AI进一步讨论。

• 将server.py MCP服务器连接到Claude桌面。可以把它想象成用Python编写自己的机器人，把内容带入聊天上下文：市场数据、SEC文件，或者任何你能检索或计算的东西。

• 提示Claude Desktop查询Perplexity、维基百科和10-K，将特斯拉信息纳入上下文中。

MCP Securities Analysis

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

Example deep research report for Tesla.

This report was generated semi-autonomously by the following steps:

• connect MCP tools to Claude Desktop, including web search, Perplexity, Wikipedia. (see claude_desktop_config.json for details). You can think of the Wikipedia tool as a friend you invite into the chat conversation who knows how to search and navigate Wikipedia, and bring articles, sections, summaries into the chat context, for further discussion with the user and the AI.

• connect server.py MCP server to Claude Desktop. Think of this as writing your own bots in Python that can bring stuff into the chat context: market data, SEC filings, or anything else you can retrieve or compute.

• prompt Claude Desktop to query Perplexity, Wikipedia, and the 10-K to bring information about Tesla into the context.

• prompt Claude Desktop to query market data tools for up-to-date 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 chat context, additional web searches and tool calls.

• see prompts.txt for prompts used in the example.

While it’s not a fully autonomous agent and at an early POC level, it shows the way toward a fully autonomous agent. Create an MCP client that autonomously goes through the steps above and generates a deep report on Tesla in a structured format with graphs and tables. Or 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.

server.py Features

• FastMCP server – server.py 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

  • yfinance
  • OpenBB. via the OpenBB platform API, we can call just about any market data REST API via a unified API.

• SEC filings - a tool gets 10-K Item 1 via sec_downloader and parses it with sec_parser.

• Technical analysis - computes a few technical indicators with pandas_ta & TA-Lib.

• Interactive plots - Make a plot with plotly