Robinhood MCP for AI Agents

Educational resource. Not financial advice. Not affiliated with Robinhood Markets, Inc. Agentic trading involves risk of loss.

Summary

Robinhood's Trading MCP (https://agent.robinhood.com/mcp/trading) lets AI agents (Claude, Cursor, ChatGPT, Codex, Grok, and other MCP clients) read Robinhood account data and place supported orders inside a dedicated Agentic account. Primary brokerage accounts stay read-only for place/cancel.

Use it as a desk-side research and agentic-execution rail — not as a drop-in substitute for a production multi-broker execution ledger.

What people use Robinhood MCP for

Portfolio and risk Q&A — balances, positions, order history, realized P&L.
Market research — quotes, fundamentals, technical indicators, earnings calendar, options chains (where eligible).
Scanners and watchlists — saved liquid-gainer / power-hour screens; organize names for a session.
Confirmed agentic equity (and eligible options) orders — review_* then place_* / cancel_* only on accounts with agentic_allowed=true.
Guarded scalp sessions — short RTH windows with hard flat times, size caps, and written entry/exit grammar (see the Agentic Equity Scalp Playbook).

Account model (agent-safe)

Read — Agent may read accounts the OAuth session exposes
Place / cancel — Only accounts with agentic_allowed=true
Funding — Agentic wallet is separately funded; that balance is the loss cap
Masking — User-facing prose should show account last-4 only; pass full account numbers from get_accounts into tools
Buying power — Use get_portfolio, not get_accounts

Tool groups

Read: accounts, portfolio, equity/option positions & orders, tax lots, realized P&L / trade history, quotes, price book, historicals, fundamentals, technicals, indexes, options, earnings
Screen / organize: saved scanners, equity/index/crypto watchlists, options watchlist
Trade: review_equity_order / review_option_order then place_* / cancel_* — confirm with the operator first unless a pre-authorized automation path is explicit
Out of scope for most clients: futures still require the Robinhood app

Agent guardrails that matter

Confirm before writes (watchlist mutate, scan create, place, cancel) unless the operator pre-authorized a bounded session.
Never invent account_number — resolve via get_accounts; never silently default when multiple accounts exist.
Skip place/cancel when agentic_allowed is false.
Prefer review tools before live place; show quotes and order preview to the operator.
Bound risk outside the model: per-trade notional, daily loss stop, hard flat clock, kill switch (disconnect MCP / cancel working orders).
Log decisions in a durable ledger the agent can re-read next session.

What Robinhood officially suggests

Robinhood's Agentic Trading overview lists example prompts such as:

Build thematic portfolios from news and industry reports
Automate simple rules (e.g. buy a fixed dollar amount after a daily drop)
Rebalance to target weights
Analyze portfolio risk and build bull/bear theses from quotes + sentiment

Those are starting points. Desks that keep expectancy usually replace open-ended prompts with a written playbook (screen → trigger → size → TP + cut → flat time). This site brands that refusal muscle as sitting on your hands — see when the agent should sit. Field notes: Evidence.

Public writeups vs field practice

Most public articles cover setup and safety (OAuth, Agentic account, per-trade caps). Fewer publish session-level win/loss grammar. This site publishes a sanitized field playbook: liquid gainers, sweep-reclaim entries, size-up checklists, and anti-patterns (MAE parked in liquidity sweeps, midday ladders without cuts, mid-reclaim fake R:R).

FAQ

What is the Robinhood Trading MCP URL?

https://agent.robinhood.com/mcp/trading — Streamable HTTP MCP. Authenticate via Robinhood OAuth on desktop.

Can an agent trade my main Robinhood account?

No. Place/cancel is restricted to the Agentic account (agentic_allowed=true). Other accounts remain read-only for this MCP agent.

What should I set before the first live order?

Per-trade notional, daily loss stop, approval vs auto-execute policy, hard flat time, and a kill phrase / MCP disconnect path.

Disclaimer

Educational only. Trading equities and options involves substantial risk of loss. AI agents can misread quotes, ignore constraints, or act on stale context. You are responsible for all agent-placed orders. This site is not a broker-dealer and is not affiliated with Robinhood.