Claude acts as financial analyst
A social thread demonstrated Claude AI running as a full financial analyst on MCP servers—pulling stock data, running backtests, and returning reproducible prompts and outputs. The post includes backtest results and prompts you can replicate, showing how LLMs are being stitched into analyst workflows today. (x.com)
Anthropic open‑sourced the Model Context Protocol (MCP) on November 25, 2024, establishing the open standard that lets LLMs like Claude query external data sources and tools during a session. (anthropic.com) Claude Code's documentation describes two MCP connection modes—interactive OAuth for live sessions and programmatic API‑key access for scripted workflows—enabling Claude to request market data or trigger backtests from remote MCP servers. (code.claude.com) The Composer MCP server project explicitly advertises the ability for MCP‑enabled LLMs to create, backtest, and execute automated trading "symphonies," and the composer‑trade‑mcp package was published to PyPI on July 14, 2025. (github.com) QuantContext's MCP package claims to turn plain‑English strategy descriptions into executable quant research with every number computed from market data and "fully reproducible" backtest outputs. (pypi.org) A Financial Datasets MCP server repository exposes tools for current stock prices, income statements, balance sheets, cash flows, and SEC filings so an LLM can fetch firm fundamentals directly rather than hallucinating numbers. (github.com) Independent writeups and demo repos document end‑to‑end setups: a May 4, 2025 walkthrough showed using Claude + MCP to query markets and validate ideas, and the open Trade‑With‑Claude cbt‑framework provides scripted commands for backtesting across multiple exchanges. (dangelov.com) Backtesting skills packaged for Claude advertise production‑grade features—vectorized backtests, explicit slippage/transaction‑cost modeling, walk‑forward analysis, and mitigation patterns for look‑ahead and survivorship biases—so outputs can be assessed as implementable research rather than casual toy results. (mcpmarket.com)
