Skip to main content

What is hoodpocket

hoodpocket is the commerce and financial layer for AI agents on Robinhood. The first product, pocket wallet, gives your AI agent a small, separate, self-custodial wallet it can trade from 24/7 on Robinhood Chain, inside limits you set.

It ships as an MCP server, so any MCP-capable agent (Claude, ChatGPT, Cursor, and others) can use it out of the box.

┌─────────────────┐ MCP ┌──────────────────────────────┐ txs ┌───────────────────┐
│ AI agent │ ─────────► │ hoodpocket │ ───────► │ Robinhood Chain │
│ (Claude, ...) │ │ classify → guardrails → sign │ │ Uniswap v4 │
└─────────────────┘ └──────────────────────────────┘ └───────────────────┘

Why it exists

Giving an agent your exchange API key or wallet seed gives it everything: keys are all-or-nothing. Prompt-level rules ("only trade $100") are suggestions the model can misread, forget, or be injected out of. And open chains are adversarial: full of tokens named after things they are not.

hoodpocket answers with three design decisions:

  1. Isolation. The pocket is a fresh wallet funded with only what the agent may trade. Worst case is the pocket, nothing more.
  2. Policy before signatures. Every trade is checked against your guardrails before anything is signed. A blocked trade costs nothing.
  3. Scam-aware classification. Every token is sorted into a trust tier using on-chain signals that cost money to fake, never its name.

What the agent can do

  • Discover tokens: memecoins (Noxa launches), utility tokens (Virtuals), tokenized stocks
  • Classify any address into a trust tier, with reasons
  • Quote and swap against native ETH or USDG through Uniswap v4
  • See the portfolio, its limits, and the full trade history

What the agent can never do

  • See the private key (there is no MCP tool that returns it)
  • Send funds out of the wallet (there is no withdrawal tool)
  • Exceed the per-trade or daily USD limits you set

Ready? Start with the Quickstart.