Dexter: Claude‑powered finance repo
An open‑source project called Dexter uses Claude to analyze undervalued stocks and auto‑generate investment theses—presented as a full‑stack portfolio idea that blends AI agents, data analysis, and web UI. The repo provides a concrete example to showcase agent integration and end‑to‑end engineering on GitHub. (social post linking Dexter repo)
The main virattt/dexter repository shows roughly 18.2k stars and about 2.2k forks on GitHub according to the project header. (github.com) The codebase is organized into separate services—named dexter-api, dexter-agents, and dexter-mcp—with dexter-mcp described as an MCP bridge and dexter-api responsible for Supabase-backed auth, prompt storage, and OAuth handlers. (github.com) Anthropic (Claude) is an explicit integration target: the README lists ANTHROPIC_API_KEY as an optional env var and the connector stack notes a default Anthropic client id used when CONNECTOR_CLAUDE_CLIENT_ID is unset. (github.com) The project requires Bun runtime v1.0+ and is TypeScript-based (tsconfig.json present), and env.example enumerates multiple optional provider keys including OPENAI_API_KEY, XAI_API_KEY, and FINANCIAL_DATASETS_API_KEY for institutional market data. (github.com) Dexter ships an evaluation suite that runs financial-question benchmarks with LangSmith for tracking and an LLM‑as‑judge scoring approach, and the README documents safety controls such as loop detection and step limits. (github.com) Active community forks and ports have proliferated (examples: techsd/dexter-agent, benbarber/dexter-ai, rushi404/dexter-ai-finance), and several walkthroughs including a “Dexter 2.0” demo video have been published on YouTube. (github.com)
