Claude can read a chart, research a company and write a trading script. Placing a real order is a different job: it needs a broker connector, your permission and a session that is still running. Here is what works in November 2026, what Anthropic allows, and how to keep the risk small.
Updated November 16, 2026
15
min to read
The short answer
Not by itself. In a chat, Claude can read a chart screenshot, research a stock and run code, but it has no brokerage account. It places an order only through a broker connector you add, such as Robinhood’s or Alpaca’s MCP server, and only while a session runs. Anthropic’s own browser and Duet agents won’t trade securities at all.
A lot, as long as the job is reading, reasoning or writing. In a normal chat, with nothing connected, Claude can:
This is a screenshot of [TICKER] on the [TIMEFRAME] chart. Describe the trend, the two nearest support and resistance levels, and what the RSI shows if it is on the chart. Then list what would prove this read wrong. Mark which numbers you read off the image and how sure you are of each one. Don’t tell me whether to buy or sell.
For more like it, see our trading prompts for charts, research and risk.
Claude is a strong general analyst: it can work through a long filing, write working code and lay out both sides of an argument. If you are choosing an assistant for this kind of work, we compare the two best-known ones task by task in Claude vs ChatGPT for trading, and answer the same question for ChatGPT in can ChatGPT trade for you? For Grok, see can Grok trade for you? Our wider LLM trading comparison sets all four, plus Gemini, side by side on the same tasks.
Out of the box, Claude has no brokerage account, no live price feed and no view of your positions. Five limits follow from that:
| Task | Claude on its own | What changes it |
|---|---|---|
| Place an order | No. A chat has no route to a broker | A broker connector that offers an order tool |
| See live prices | Only what its training data or a web page says | A connector to a broker’s or a data provider’s feed |
| Hold your account or keys | No. It has no login or wallet of its own | A connector you sign in to, or API keys in a config file |
| Watch a market with the chat closed | No | A Claude Code scheduled task, at most once an hour |
| Keep a stop in place | No. A stop that only exists in a chat is gone when the chat ends | A real stop order resting at your broker |
Anthropic draws the same line in its own agent products. Claude in Chrome, the extension that clicks through websites for you, lists “Executing financial trades or investment transactions” among the actions it is prohibited from taking, whatever permissions you give it. Claude Duet, the agent app Anthropic opened as a research preview on October 21, 2026, keeps working when you step away. Yet Anthropic writes: “A few things are off-limits even with your approval, like trading securities or moving money between accounts.” So asking Claude to log in to your broker’s website and buy shares will not work.
What about watching a position? Claude Code, Anthropic’s coding agent, can run a saved prompt on a schedule in Anthropic’s cloud, even with your laptop closed. Its docs reject schedules that run more often than once an hour, and say the saved prompt “can’t act as approval or consent for actions during the run.” That is fine for a morning summary. It is no substitute for a stop.
Through MCP, the Model Context Protocol: an open standard that lets an AI app use outside tools. A broker runs an MCP server, a service that offers tools such as “get positions” or “place order.” You add it to Claude, sign in, and Claude can call those tools when you ask. There are three common setups:
claude mcp add. Traders who code use it to combine broker tools with their own scripts.When Claude wants to use a tool, it shows you an approval request. Anthropic’s connector guide says to “only click ‘Allow always’ when using a server and tool that you trust to run unsupervised.” For a tool that places orders, the safe answer is not to.
| Connector | Made by | What it can do | Worth knowing |
|---|---|---|---|
| Robinhood Trading MCP | Robinhood | Reads your accounts; equities, options and crypto tools | Trades only in a separate agentic account you fund |
| Alpaca MCP server | Alpaca, open source | Market data and orders for stocks, options and crypto | Starts in paper trading; live trading takes a setting change |
| TradingView MCP server | TradingView | Quotes, screeners, technical ratings, news and alerts | Places no orders |
Paper trading means simulated orders with pretend money against real prices. Alpaca’s server starts there: its README sets ALPACA_PAPER_TRADE to true by default and tells you to set it to false for live trading. Its security notice is plain: “This server can place real trades and access your portfolio.” If you only want Claude to read market data from TradingView, see how to connect Claude to TradingView with MCP.
There is a fourth route, the one people mean by a “Claude trading bot”: asking Claude Code to write a bot that calls a broker’s API and runs on a server. Then your code trades, and Claude is only the programmer. The bot does exactly what the code says, bugs included, so run it on paper first. If someone offers to sell you a ready-made one, read is AI trading legit? before you pay. A “Claude trading skill” is something else again. Skills are folders of instructions, scripts and resources that Claude loads for a specialized task. A skill can teach Claude a method, but connecting it to a market is the job of an MCP connector. Our guide to building a Claude trading bot with Claude Code covers that route in full, brokers, keys and what goes wrong included.
Yes, through Robinhood’s agentic trading, which Robinhood launched in May 2026 for customers who want to connect their own AI agents. The setup, from Robinhood’s support page:
In Claude, add a custom connector with the URL https://agent.robinhood.com/mcp/trading. In Claude Code, run claude mcp add robinhood-trading --transport http https://agent.robinhood.com/mcp/trading, then type /mcp and pick robinhood-trading.
Authenticate with Robinhood and you are prompted to open an MCP account, a separate account just for your agent. You need a primary individual investing account in good standing.
Robinhood says your agent “can only trade with funds you deposit into this account.” The deposit is your real limit.
Trade approvals are off by default for MCP accounts, so turn them on in the account’s Safety Controls if you want to place each order yourself. Robinhood warns that if you’ve asked it to act without your approval, “it can place trades without your confirmation.”
Two lines on Robinhood’s page are worth reading twice. First, the agent gets read access to “all your Robinhood accounts, including your Robinhood account numbers,” plus your positions, balances and order history, even though it “can only place trades in your Robinhood Agentic account.” Second, the responsibility stays with you: “You are ultimately responsible for the trades your AI agent places in your account,” and AI agents “can make errors, misinterpret instructions, act on incomplete or outdated information, and may behave in unexpected ways.”
Expect some friction too. Since late August 2026, some traders have reported that Claude declines to place orders through Robinhood’s connector while others say theirs still works, according to a detailed user report. We found no rule from Anthropic about trades through connectors, but its usage policy says its products include “real-time safeguards that may block or limit certain outputs.” If Claude reads your account but won’t place orders, first check that the agentic account is open and funded.
Robinhood is also building agents into its own app. On September 29, 2026 it announced Robinhood Agents, coming to eligible US customers: you pick a model from one of several AI labs inside the Robinhood app, so there is no connector to set up. We look at them, and at the alternatives, in Robinhood Agents and its alternatives.
Five risks come up again and again. None is unique to Claude: any AI model wired to an account carries them.
| Risk | What it looks like | How to limit it |
|---|---|---|
| Confident mistakes | A wrong level read off a screenshot, a misread instruction, an order for the wrong size | Keep approvals on and read every order before you allow it |
| Stale data | An answer built on training data or a delayed quote instead of the live price | Ask where each number came from and when; check the price at the broker |
| Prompt injection | Hidden text in a web page, a news story or a tool reply that tells Claude to do something you never asked | Keep trading sessions away from untrusted pages; connect only servers you trust |
| Keys and permissions | An API key saved in a config file, a connector that can read all your accounts | Use paper or a small separate account; remove connectors you no longer use |
| No stop when the chat ends | A plan to “sell if it drops 5%” that only exists in a conversation | Place the stop as a real order at the broker |
Prompt injection deserves a closer look, because it turns reading into acting. It means instructions hidden in content the model processes. Anthropic’s own research on browser agents says “prompt injection is far from a solved problem, particularly as models take more real-world actions,” and that even “a 1% attack success rate” still “represents meaningful risk.” Its connector guide adds that malicious MCP servers “may include hidden instructions that try to make Claude perform unintended actions.” The session to keep separate is the one that can both read the web and place orders.
Run the whole flow with simulated money first, for example on Alpaca’s paper account, until you trust every step.
Trade from an account that holds only what you could lose, like Robinhood’s agentic account.
Never choose “Allow always” for an order tool. Read the symbol, side, size and order type before you allow it.
Send stops and limits as real orders, so they stay in place when the chat closes.
Turn off the tools a chat doesn’t need, and don’t mix trading with browsing.
Start each session with your limits: the largest size you’ll trade, the markets you trade and when to stop for the day.
Anthropic’s Usage Policy, in the version effective November 12, 2026, treats finance as a high-risk area. It defines the case as “a personalized recommendation to buy, sell, hold, or allocate among specific investment products, such as securities, funds, digital assets, or annuities.” When Anthropic’s products are used to make that kind of recommendation, two controls apply:
The policy also lists what falls outside those requirements. That includes general or educational content, such as explaining “how investing works” without applying it to your circumstances; “internal drafting, research, summarization, or analysis”; and “carrying out a decision a person has already made.” For agents, users are responsible for making sure the agents they build or deploy comply, “including actions those agents take through tools, browsers, or connected systems.”
Read it as a line between two kinds of use. Research, explanation and carrying out a decision you already made are listed outside the requirements. Personalized buy or sell calls delivered to people who rely on them need a qualified person reviewing them and a clear AI label. The policy adds that running into one of Anthropic’s real-time blocks “does not by itself indicate a violation.”
Claude is a general assistant you can wire to a broker yourself. An AI trading agent is built for the job from the start: it comes with market data, an account to trade from and rules that keep working after you close the tab. Ethy is one: the AI trading agent you ask before you trade. There is no broker to connect and nothing to install.
Ethy also does the trading side. From the chat it places orders on Hyperliquid perpetuals, or perps: contracts that track a price with no expiry date and can use leverage, which magnifies losses as well as gains. They cover crypto, plus stocks, indices, commodities and FX listed on trade.xyz, an exchange that runs on Hyperliquid, and Ethy also runs spot swaps. Whether a chat order shows you a confirmation card first is a setting you choose. Automations trade on their own once you activate them, and paper mode tests one with live prices and no money. Stop loss and take profit orders sit on Hyperliquid itself, so they fire even when Ethy is closed. Every account gets a wallet at sign-up: your balance lives in a wallet that belongs to you, and you can export its key at any time.
Claude remains the better tool for open-ended work beyond trading: long documents, software, anything that is not a market. Ethy is narrower, and it covers the trading jobs from the first question to the order.
Only through a connector to an account that can trade crypto. Robinhood’s Trading MCP includes crypto tools for its agentic account, and Alpaca’s MCP server supports crypto orders, starting on paper. Claude itself holds no coins and no keys. Crypto trades around the clock, so a position can move while no chat is open: keep a real stop order at the exchange or broker.
Claude is a real assistant from Anthropic, and broker connectors such as Robinhood’s are real products. But Anthropic built its own agents so they won’t trade securities, so anything sold as a “Claude trading bot” is made by someone else. Judge it like any third-party trading software, test it on paper and walk away from anyone who promises returns.
Robinhood’s Trading MCP lists options tools next to its equities and crypto tools, so a connected Claude can work with options inside your agentic account. Options can lose their whole value quickly. Keep trade approvals on and check the strike and the expiry before you allow an order.
For Claude to place an order, a session must be running: a chat, a Claude Code session, or a scheduled task that runs at most once an hour. Orders already resting at your broker, such as a stop order, stay there whether Claude is open or not. That is why your protection belongs at the broker, not in the chat.
A skill is a folder of instructions, scripts and resources that Claude loads for a specialized task, such as an earnings checklist or a script that computes an indicator. A skill doesn’t connect Claude to a market or a broker; MCP connectors do that. Read a skill’s scripts before you run it, as you would with any code.
Yes, in code. Upload price data and Claude can write and run a backtest in its sandbox, or build a fuller one in Claude Code. The result is only as good as the data and the code, so check how it handles slippage and look-ahead bias, which is using information a trader could not have had at the time.
Trading involves risk, including the loss of your money. Ethy AI is a research and trading tool and does not give investment advice or recommendations. Backtests are simulations on historical data and do not predict future results.
Deploy your AI trading agent in one click.
