How to set up Robinhood agentic trading with Claude Code

The one-sentence map of this ecosystem: Robinhood provides the execution connection. Claude provides the agent. Coil provides the deterministic trading strategy — either served to your agent over MCP ($29/mo, nothing installed) or downloaded and owned ($79 once). This page covers where each piece begins and ends.

This is the setup walkthrough we wished existed: five steps from "I've heard Robinhood lets an AI trade now" to a connected, contained, supervised agent — written to be useful even if you never buy anything from us. The five steps: understand what Robinhood's agentic account actually is, open and fund it with a deliberately small balance, connect Claude Code over Robinhood's MCP, arm the safety guardrails before any trading, and give the agent a rule-driven strategy. Most tutorials get you to step 3 (the connection) and stop. The connection is the easy part. Steps 4 and 5 — the safety rails and the strategy — are where the outcome is actually decided, so this guide spends its weight there. If you only read one section, read step 5: a connected agent with no strategy trades headlines, and the three honest ways to fix that are laid out there.

Coil itself is agent-native on the same pattern — the same claude mcp add move you will make in step 3: a free MCP server for delayed market reads at claude mcp add --transport http coil https://coil.trade/mcp, free API keys (POST https://coil.trade/api/key), pay-per-call x402 endpoints, and — the one that answers step 5 — a Coil Live subscription that unlocks the live engine-grade feed and the served rulebook on that same server, by licence header. Full docs on the For AI agents page; the read order that pairs Coil's selection with Robinhood's execution is at /agents/robinhood.

Companion references as you work through it: the complete 53-tool MCP reference (what your agent can actually call, with measured gotchas), the server-options comparison if you're weighing the official connector against open-source bridges, the error-by-error troubleshooting table when something refuses to work, and the open reference architecture on GitHub for keeping the agent outside the trading-decision boundary.

Coil is independent and is not affiliated with, endorsed by, or partnered with Robinhood Markets, Anthropic, or OpenAI. Everything below about their products comes from their public materials as of July 2026 — features, availability, and pricing change, so verify the current details on their own sites. Nothing on this page is investment advice or a recommendation to buy, sell, or hold any security.

Step 1 — Understand what Robinhood agentic trading actually is

Robinhood launched agentic trading on May 27, 2026, in beta — equities-only at launch, rolling out account by account (as of this writing; the current status lives on Robinhood's official agentic-trading page). The design is more careful than the headlines suggested, and it's worth understanding before you touch it:

• A dedicated agentic brokerage account. The agent doesn't get your main portfolio. It trades inside a separate account with its own dedicated balance — a budget you set by deciding what to transfer in.
• Read-wide, trade-narrow. The agent can read portfolio context to inform decisions, but it can only place trades inside the agentic account. Your long-term holdings are outside its reach.
• A push notification per trade. Every order the agent places pings your phone. You find out in real time, not at month-end.
• A live activity feed. The agent's activity and performance are visible in the Robinhood app as they happen.
• One-tap disconnect. You can sever the agent's access directly in the app, anytime, without touching your agent's configuration.

One line from Robinhood matters more than all the marketing: it "does not control, supervise, monitor, recommend, or audit" the AI agents that connect. Robinhood built the rails — the account isolation, the notifications, the kill switch. What runs on those rails is entirely your problem. That single fact shapes steps 4 and 5 of this guide, and it's the honest reason a page like this needs to exist.

Step 2 — Open and fund the agentic account

In the Robinhood app, agentic trading is enabled from your account settings (the exact placement has moved during the beta — Robinhood's page and in-app search will find it). Because the rollout has been account by account, you may see a waitlist rather than an instant toggle; that's normal for the beta.

Then fund it — and this is the first real decision of the whole setup: your deposit is your risk dial. The agentic account's balance is the hard ceiling on what any agent, any strategy, or any bug can lose. So set it like an engineer, not an optimist:

• Start with an amount you could lose entirely and shrug off. Not "probably won't lose" — could lose, to zero.
• You can always transfer more in after weeks of watching. You cannot retroactively shrink a loss.
• Don't fund it with money that has a job (rent, emergency fund, tuition). This is risk capital by definition.

Everything downstream — the MCP connection, the strategy, the automation — operates inside this boundary. Getting the boundary right first is what makes the rest of the setup low-stakes to learn on.

Step 3 — Connect Claude Code over Robinhood's MCP

MCP (Model Context Protocol) is the standard interface that lets an AI agent call external tools — in this case, Robinhood's own brokerage tools: read positions, get quotes, review and place orders. Robinhood publishes the connection as "paste one URL": you take the MCP URL from its official agentic-trading page, add it to your agent's MCP configuration, and authenticate.

At launch, Robinhood's supported agents included Claude Code, Claude Desktop, ChatGPT (developer mode), Codex, and Cursor — the current list is on Robinhood's page. The flow is essentially the same everywhere:

1. Get the MCP URL from Robinhood's official page. Only from there. Do not copy an MCP URL out of a YouTube description, a Reddit comment, or a third-party blog — a wrong URL is how credentials get phished. We deliberately don't print the URL here for the same reason: get it from Robinhood, where it's guaranteed current.
2. Add it to your agent's MCP config. In Claude Code, MCP servers are added via the CLI (claude mcp add) or its settings; in Claude Desktop, via the connectors/settings screen. Each agent's docs cover its own two-minute version of this. One flag trips people up: a bare claude mcp add <name> <arg> registers a local command, so a remote endpoint like Robinhood's needs the HTTP transport named explicitly.

   claude mcp add --transport http robinhood-trading <URL from Robinhood’s agentic-trading page>

3. Authenticate via OAuth. Your browser opens a robinhood.com login page; you sign in and approve the connection. This is the critical security property: you never type your Robinhood password into the agent, into a config file, or into any third-party site. If any tutorial or tool asks for your Robinhood password directly, close it. The OAuth page should be on robinhood.com — check the address bar.
4. Verify with a read before any write. Ask the agent something harmless: "list my positions in the agentic account" or "quote AAPL." A clean read-only round trip confirms the connection works before an order is ever on the table.

Connected everything and the agent still refuses to trade? That has five distinct causes that all look identical from the chat window — work through the ordered diagnostic before touching your prompt.

That's genuinely all the connection is. If a tutorial makes this step look complicated, it's padding. The complicated part comes next.

Step 4 — The safety guardrails (arm these before any trading)

Two layers, and only one of them is Robinhood's job. The broker-side guardrails come free with the agentic account — the balance you funded in step 2 is the hard ceiling, plus a push notification per trade, a live activity feed, and one-tap disconnect. The your-side guardrails nobody supplies: the rules the agent cannot improvise or argue its way around. Step 5 is that second layer; this step arms the first.

The copy-paste version: a ready guardrail template covering all of this — account lockdown, order caps, loss halt, one-and-done, kill switch, and the prompt-injection rule — lives at coil.trade/guides/guardrails.md. Paste it into your agent's standing instructions and edit the CAPS values. It is the generalized version of the rules our own engine runs live.

Robinhood gave you containment tools. They only work if you arm them before the first trade, not after the first surprise. The guardrails:

• Keep the balance small until trust is earned. The deposit from step 2 is the only cap that cannot be talked around, prompted around, or bugged around. Weeks of watching first; scale later, if ever.
• Turn trade notifications on and test them. You want to know what an agent-trade push notification looks and sounds like on a calm day, so an unexpected one on a volatile day registers instantly.
• Find the disconnect switch now. Open the app, locate the one-tap disconnect for the agent connection, and remember where it is. In a moment where you actually need it, you don't want to be searching menus.
• Never share credentials with third-party servers. The OAuth flow in step 3 is the only legitimate credential path. Any tool, bot, "signal bridge," or hosted service that wants your Robinhood login or an API token pasted into their system is a hard no. (This is also a fair test to apply to us: Coil runs locally on your machine and your credentials never pass through any server of ours.)
• Know what leaves Robinhood. Robinhood is explicit that account data shared with your chosen AI provider exits its security environment. That's inherent to agentic trading with any hosted model — decide you're comfortable with it before connecting, not after.
• Don't automate unattended on day one. A scheduled agent session that trades while you sleep is the end state, not the starting point. Run sessions manually and watch them until the agent's behavior is boring. Boring is the goal.

Step 5 — Give the agent a strategy (the step every tutorial skips)

Here is the moment every setup video ends on: the MCP is connected, the agent lists your positions, everyone celebrates. But notice what you actually have — an agent that can trade, with no idea what to trade or why. Unprompted, a large language model with a brokerage tool will do what it always does: improvise something plausible. It will trade headlines. It will trade vibes. It will buy a name because it was in the news and sell it because a later prompt sounded worried — with no consistent entry logic, no exit plan, no position sizing, and no memory of why it did what it did last week. Robinhood explicitly doesn't supervise the agent, so nobody is checking that work but you.

You have three honest options, and they're genuinely different amounts of work:

• Prompt it ad-hoc. "Look at my watchlist and tell me if anything's worth buying today." Fine for learning the tools, and the per-trade notifications keep you in the loop. But you are the strategy, the risk manager, and the discipline — every session, forever. Consistency is exactly what humans prompting off the cuff are worst at, and it's fragile in the specific way that costs money: the rules drift with your mood.
• Write your own rule set. The real version of this is weeks of work, not an evening: define a universe, entry conditions, exit conditions, position sizing, stop placement, drawdown halts — then test it against history without fooling yourself, which is its own discipline (see how to read a backtest and survivorship bias, explained for the two classic self-deceptions). If you enjoy that work, it's the most educational path there is, and you should take it.
• Point it at a finished, rule-driven system. This is the category Coil is in, so read the next two paragraphs knowing we sell the thing being described. It splits into a hosted feed your agent calls, and a system you download and own.

The hosted path, because you just did the hard part. You have an agent wired to a broker over MCP. Coil Live ($29/mo or $249/yr) is a second MCP server, added the same way: your agent calls https://coil.trade/mcp with your licence key as a header, and gets back a served entry list — the names whose entry window reads READY right now, each with its structural stop, target, risk-reward, conviction and lane — plus a versioned rulebook it reads and follows (cycle order, exits before entries, sizing precedence, PDT bounds, and a fail-closed rule for stale data). It is the same board and the same discipline behind the live record, refreshed every 5–9 minutes in market hours. Nothing to install, no market-data account, no second credential path to your broker: your agent still places every order itself through Robinhood's MCP from step 3, and Coil never sees your account. The served rules start any agent in dry-run until you deliberately tell it otherwise. An empty list on a given day is the discipline working, and the payload says so.

The owned path. Coil Engine is a $79 one-time system your agent runs: a scanner that scores every name in the S&P 500, Nasdaq-100, and a macro book (bonds, gold, income, commodities) for opportunity, entry window (READY / SETUP / WAIT / CHASE / FALLING), hold-conviction, leadership, and sector rotation; a local dashboard; and a long-only, rule-driven engine on top. The rules are the point: it never buys FALLING names, never CHASEs extended ones, holds structural stops 4–14% below entries at volume-profile and fib levels, and treats a downturn as cash plus defensive macro rotation — never shorts, never inverse ETFs. When a leader is worth accelerating it may map to a leveraged ETF at reduced notional (NVDA→NVDL, QQQ→TQQQ) — and those vehicles can lose value rapidly, including total loss, which is why the notional is reduced and the decay mechanics are documented. Two facts matter for this guide specifically: the Engine is software your Claude runs locally, driving Robinhood's own MCP from step 3 rather than opening a new path to your money (Coil Live is an MCP server, but a read-only one — it serves data and rules, and cannot reach a brokerage); and both ship with live trading OFF, so the honest first week is watching it score and decide in dry-run against the same market you're reading about.

The numbers behind that option, framed honestly

These are research-backtest figures — not live results, not client returns, and not a prediction. The leadership-rotation backbone Coil's scoring is built on was replayed point-in-time and survivorship-free (delisted names included), with next-open fills and costs modeled. Every number rides next to its same-window SPY benchmark.

In that research backtest the backbone compounded +638% versus SPY's +282%, with a shallower worst drawdown (−23% vs −32%) and a positive result in 9 of 10 years (worst −1%, in 2018). Read the honest rider with it: through the end of 2025 it ran roughly even with SPY at about one-third less drawdown — the outperformance concentrates in leadership regimes (2025 +51%, 2026 H1 +86%). The engine itself is newly live; these are research figures, not a live track record, and past performance does not predict future results.

Whichever of the three options you choose, choose one deliberately before the agent trades. The worst outcome in this category isn't picking the wrong strategy — it's connecting an agent, skipping this step, and discovering weeks later what "no strategy" traded like. For how the pieces fit conceptually — scheduled sessions, the safety spine, what the agent does versus what the rules do — the companion guide how to run an AI agent that trades on Robinhood goes deeper; for the literal text to paste — a copy-paste prompt that encodes the read order, regime first and the broker's tools last — see the Coil × Robinhood pairing page; and if you're weighing packaged systems against bots and paid alert services generally, that comparison is here.

FAQ

Is Robinhood agentic trading safe?

Split the question in two. The rails are deliberately contained: a dedicated account separate from your main portfolio, a push notification per trade, a live in-app activity feed, one-tap disconnect. That part is well designed. The strategy is entirely unsupervised: Robinhood does not control, monitor, or audit the AI agents that connect, and its own risk language warns you can lose your entire investment. So "safe" is something you construct — small dedicated balance, notifications armed, rules the agent can't improvise around — not something the platform hands you. No configuration makes trading risk-free.

Can Claude trade stocks for me?

Mechanically, yes: Claude Code connected to Robinhood's MCP can read positions, get quotes, and place equity trades inside your agentic account. Practically, the mechanical part is the least important part — Claude arrives with no trading strategy, and an unprompted agent trades headlines and improvisation. What Claude is genuinely excellent at is operating a fixed rule set with perfect consistency: running the same checklist every session without boredom, fear, or greed. Give it the rule set; that's step 5.

Do I need to know how to code?

Not for the setup in this guide. The connection is pasting a URL and approving a browser login. Where code (or weeks of very careful prompt engineering) enters is writing your own strategy — universe, entries, exits, sizing, stops, testing. If you'd rather not, a packaged system covers that layer: Coil's install is agent-led — you unzip it, point Claude Code at the folder, and the onboarding handles dependencies and the free data key interactively. No code is required from you at any step.

What does it cost?

As of July 2026 (check each site for current terms): the Robinhood account is free to open, with commission-free equities trading and standard regulatory fees. Your AI agent runs on either a paid Claude plan or pay-as-you-go API billing — a scheduled swing routine is a handful of tool calls a few times a day, which fits a modest plan; an always-on intraday watcher is what burns limits. Vendor pricing is Anthropic's to set and change. Coil, if you choose the finished-system route, is either $29/mo (Coil Live — the hosted feed and rulebook your agent calls over MCP, nothing installed) or $79 one-time (Coil Engine — the whole system as source on your machine, no subscription). The number that matters most isn't a price at all: it's the balance you put in the agentic account, because that — not any fee — is what's actually at risk.

What prompt should I give Claude once it's connected?

Yes, there is a pre-written version — but it is a read order plus your own risk limits, not a magic phrase. An agent told "trade my account profitably" screens whatever it happened to see last; an agent told to check the market regime first, take candidates only from a ranked list, and size from rules you wrote behaves the same way every session. The copy-paste version, with the read order it depends on, is on the Coil × Robinhood pairing page. If you subscribe to Coil Live, the prompt collapses to one line — “call get_live_setup and follow it” — because the read order, the cadence and the rulebook are served to the agent rather than typed by you. Two things stay in your half of that prompt and never the publisher's: position size and exit rules. Coil publishes scores and states — never position sizes, stop prices or target prices.