Skip to main content
The guardrails are set before your agent is even created. You define exactly what it can do, what it can spend, and which wallets it can touch - and Shuriken enforces all of it server-side. The agent can’t override any of it, no matter what. This page explains each permission in detail, how trading limits work, and recommended setups for different use cases.

​
Permissions overview

There are six permissions you can toggle when creating a custom Agent Key. Each one unlocks a specific set of capabilities.
Permissions are enforced on the server side. Even if an AI agent tries to bypass its restrictions, Shuriken’s server will reject any unauthorized requests. Your funds are protected regardless of what the AI tries to do.

​
Read Wallets

View your connected wallet addresses and names.

​
Read Balances

View token balances, holdings, and profit/loss data.

​
Read Alpha

Access token mentions from your connected Discord/Telegram groups and trending calls from X (Twitter).

​
Read Tokens

Look up token metadata, market data, and trading statistics.

​
Get Quotes

Create trade plans and get price quotes without actually executing. Useful for research and price checking.
The Get Quotes permission gives the agent plan_trade access. This creates a trade plan with a 5-minute expiry, but the agent cannot execute it without the Execute Trades permission. Think of it like window shopping - you can check the price tag but can’t buy.

​
Execute Trades

Actually move real funds - buy and sell tokens from your wallet.
This is the only permission that can move money. When enabled, additional safety controls (trading limits) become available. We strongly recommend setting conservative limits, especially when starting out.

​
Trading limits

When you enable Execute Trades, you get access to fine-grained controls that cap what your agent can do. These are enforced server-side - the agent cannot override them.
Trading limits configuration showing buy, sell, and execution controls

​
Buying limits

​
Selling limits

​
Execution & safety

​
Additional options

​
How the two-step trading flow works

Trading always happens in two steps. This is a safety feature - the agent must get a quote before it can execute. Step 1: Plan - The agent calls plan_trade to get a quote. The response includes:
  • Exact input and output amounts
  • Estimated USD value
  • Slippage and price impact
  • A plan ID that expires in 5 minutes
  • A policy check showing remaining limits (e.g., “8.64/8.64 / 500.00 daily buy”)
Step 2: Execute - The agent calls execute_trade with the plan ID. The server validates the plan hasn’t expired and all limits are still within bounds, then submits the transaction.
If a plan expires (after 5 minutes), the agent must create a new one. This prevents stale quotes from being executed at outdated prices.
Here are some common configurations depending on what you want your agent to do:
Permissions: Read Alpha, Read TokensUse case: You want the AI to help you research tokens and monitor alpha calls without seeing your wallet or having any trading access.Risk level: None - the agent can’t see or touch your funds.
Permissions: Read Wallets, Read Balances, Read Tokens, Get QuotesUse case: You want to ask the AI about your holdings, check PnL, and get trade quotes without giving it the ability to execute.Risk level: None - the agent can see your portfolio but can’t make any changes.
Permissions: All six enabledTrading limits:
  • Max $10 per trade
  • Max $50 per day
  • Max sell size: 25%
  • Max slippage: 5%
Use case: You want the AI to be able to trade small amounts while you learn how it works.Risk level: Low - even if something goes wrong, exposure is capped at $50/day.
Permissions: All six enabledTrading limits:
  • Max $100 per trade
  • Max $500 per day
  • Max sell size: 100%
  • Trades per hour: 20
Use case: You’re comfortable with the agent trading and want it to have enough room to act on opportunities.Risk level: Medium - higher limits mean more exposure if the agent makes a bad call.

​
Security best practices

  • Start read-only. Create a read-only key first to test the connection and get comfortable before enabling trading.
  • Use separate keys. Create one key for research and a different key for trading. This way you can revoke trading access without losing your research setup.
  • Set conservative limits. Start with low dollar limits and increase them only after you’re confident in your setup.
  • Review activity regularly. Check the Agent Keys dashboard to see every action your agent takes in real-time. If anything looks off, pause or revoke the key immediately.
  • Revoke unused keys. If you’re not using a key anymore, revoke it. You can always create a new one.
  • Never share your key. Your Agent Key is like a password. Don’t post it publicly, commit it to Git, or share it in Discord.

​
Next steps

Ready to connect your agent? Pick your AI tool:

OpenClaw

Cursor

Claude Desktop

Windsurf

Other MCP Clients