Is Veranta MCP (formerly Avantis) safe to connect to an AI agent?
Approval is optional. Safety 75%
What an AI agent can do
- Live vs paper: live on Base; testnet paper trading with test USDC source
- Order approval: Official hosted MCP (36 tools): a read-only server at https://mcp.veranta.xyz/mcp with no login, and a trading server at https://mcp.veranta.xyz/mcp/trade where your wallet signs once to create a session key (30 days by default) that can open, close and modify positions but can't withdraw USDC from your account, grant allowances or add delegates; a policy you sign caps leverage, size, daily notional, open positions and daily loss inside the tools (hosted default: 3x, $250 per trade, $100 daily loss) and the AI can't edit it; every trade is quoted first, but a confirmation is required only above a notional you set (off by default), so per-trade approval depends on your AI client; pause_trading kill switch; the local npm server is not published yet. Veranta is the renamed Avantis perpetuals exchange on Base (AVNT token) source
- Markets: perpetuals on crypto, US stocks and indices, FX, gold and oil
- Pricing: no MCP fee stated; Veranta trading fees, spread and funding apply; builder fees up to 1% of notional per order side only against an allowance you grant
Safeguards that exist
- Can it place live orders unattended?: Every trade is quoted first, but a confirmation is required only above a notional you set (off by default) and for vault moves; otherwise whether you approve each fill depends on your AI client's tool prompts and the conversation source
- Order approval or confirmation options: Approval/confirmation step documented source
- Paper, demo or testnet mode: Testnet mode (--network=testnet) trades test USDC from a faucet, so nothing is real money (official docs) source
- Withdrawal permission needed or allowed: The agent's session key can't withdraw USDC from your account, grant allowances or add delegates; margin it withdraws only pays your own wallet (official docs) source
- Scoped keys, permission scopes or IP whitelisting: Scoped access (scopes, read-only mode, dedicated or separate account) source
Safeguards that are missing
- Open source or audited: No public source code in our listing source
Safety score
| Input | From our listing | Points |
|---|---|---|
| Can it place live orders unattended? | Every trade is quoted first, but a confirmation is required only above a notional you set (off by default) and for vault moves; otherwise whether you approve each fill depends on your AI client's tool prompts and the conversation source | 10 / 25 |
| Order approval or confirmation options | Approval/confirmation step documented | 15 / 15 |
| Paper, demo or testnet mode | Testnet mode (--network=testnet) trades test USDC from a faucet, so nothing is real money (official docs) source | 15 / 15 |
| Withdrawal permission needed or allowed | The agent's session key can't withdraw USDC from your account, grant allowances or add delegates; margin it withdraws only pays your own wallet (official docs) source | 15 / 15 |
| Scoped keys, permission scopes or IP whitelisting | Scoped access (scopes, read-only mode, dedicated or separate account) | 10 / 10 |
| Open source or audited | No public source code in our listing | 0 / 10 |
| Sourced and verified facts | Facts from the vendor's official docs | 10 / 10 |
| Total | 75% of the points we could check | 75 / 100 |
Sources
- https://sdk.veranta.xyz/ai/overview
- https://sdk.veranta.xyz/ai/safety
- https://sdk.veranta.xyz/ai/session-keys
Full listing: Veranta MCP (formerly Avantis) ยท How the score works: Safety score method