Robinhood Lets Your Claude or ChatGPT Agent Trade Stocks for You — In Beta to 27M Customers

Robinhood opened its brokerage platform to autonomous AI agents on May 27, letting Anthropic’s Claude, OpenAI’s ChatGPT or any other LLM that speaks Model Context Protocol read a user’s portfolio, analyze risk and place real equity orders — the first time a major US brokerage has handed agents the keys to retail accounts. The feature, called Agentic Trading, is in beta and goes live across Robinhood’s 27 million customers, according to TechCrunch, CNBC and Bloomberg.

The shape of the product is deliberately conservative. Customers create a separate sub-account for their agent, fund a sandboxed wallet, and the agent can only spend what is inside it. Trades stream into the regular Robinhood app as notifications, and the system can be configured to require human approval on high-risk orders. “We’ve heard a lot of demand from our customers to bring their own tools, LLMs and agents, and connect them to Robinhood,” vice president of product Abhishek Fatehpuria told TechCrunch.

Under the hood, the integration runs through MCP — the open protocol Anthropic released in late 2024 and which has since become the connective tissue between LLMs and SaaS APIs. From an agent’s side, Robinhood is just another MCP server with verbs like analyze_concentration, read_analyst_notes and place_order. From Robinhood’s side, it is a way to outsource the hard problem of “which LLM” to whichever provider customers already pay for. A fraud-detection team reviews suspicious agent trades, and the company says it will help customers dispute orders an agent gets wrong.

The roadmap is the real story. Equities are live today, but Robinhood says options, crypto, event contracts, futures and prediction markets are all slated to follow “soon.” At that point, an MCP-connected Claude or GPT would be able to run an entire speculative book — hedging an NVDA position with puts, parking dry powder in stablecoins, and laying a leveraged bet on a Kalshi contract — all out of one wallet, with one set of guardrails. Alongside trading, Robinhood also launched a virtual agentic credit card for Gold members, letting an agent make purchases against a separate monthly limit; the same wallet-of-things model, applied to spending.

Regulators have not yet weighed in publicly, and Robinhood is being careful to frame Agentic Trading as a user-directed tool rather than discretionary advice — the agent acts on behalf of the customer, with a hard wallet boundary and visible audit trail. But there is no real precedent: FINRA rules were written for human brokers and algorithmic order routing, not for a probabilistic copilot that decides on its own to rebalance into Supermicro. The next twelve months will tell us whether 27 million retail traders — and the SEC — are comfortable with a model where the trader and the algorithm are the same entity.