Viral Claude trading setup exposed
A Chinese developer’s posted tutorial accidentally revealed a Claude‑powered agent setup that reportedly generated about $868K through tens of thousands of BTC trades, sparking viral interest and copytrading experiments that turned small stakes into quick profits. Other users have since tested similar Claude scripts overnight and reported copytrade gains on small capital, while bench posts showed Claude Opus accuracy slipping on coding tasks compared with peers. (x.com, x.com, x.com)
A how-to post about wiring Claude into a crypto trading stack spilled into public view, and traders on X quickly started trying to copy it with real money. (github.com, x.com) The setup uses Model Context Protocol, or MCP, a standard Anthropic introduced in November 2024 so Claude can call outside tools instead of only returning text. In practice, that means Claude can read chart data, check trading rules, and send orders through an exchange connection. (anthropic.com, claude.com) A GitHub repository posted in early April says it connects Claude Code to TradingView and executes trades automatically through BitGet, with a rules file, a safety check, and a trade log. The README says the bot can run on a schedule in the cloud and records each trade in a comma-separated values file for accounting. (github.com) That matters because brokerages and trading platforms are now publishing consumer guides for Claude-based execution, not just research. Public.com said on March 25, 2026, that Claude Desktop can place stocks, options, and crypto orders through its own MCP server, with order execution going directly from a user’s computer to the brokerage. (public.com) The result is a new kind of retail automation loop: a language model writes or runs the logic, outside tools provide the market data, and the connected account places the trade. The same architecture that helps with coding or spreadsheets can now touch a brokerage balance in one conversation window. (claude.com, public.com) The social-media claims around profits are harder to verify than the underlying plumbing. The X posts cited in the viral thread were not readable through this reporting interface, so the reported dollar gains and trade counts could not be independently confirmed from the original posts here. (x.com, x.com) What can be confirmed is that the copycat ecosystem formed fast. New repositories, videos, and tutorials describing Claude-linked automated trading appeared within days, including guides for BitGet, TradingView, and other brokerage connections. (github.com, youtube.com, quantlabsnet.com) At the same time, benchmark chatter pulled the story in the opposite direction. Anthropic’s current Opus page says Claude Opus 4.6 scores 80.8% on SWE-bench Verified, a slight dip from the 80.9% Anthropic reported for Claude Opus 4.5 in November 2025, even as Opus 4.6 improved on other coding and agentic tests. (anthropic.com, anthropic.com) So the thread that went viral was not just about one trader’s screenshots. It landed at a moment when the tools to let Claude act on markets are spreading quickly, while arguments over how much to trust the model’s judgment are still playing out in public benchmarks and live accounts. (anthropic.com, public.com, anthropic.com)
