How much does it cost to let an AI agent swap crypto?

Discovery, quotes and token-safety checks are free and need no key. A free self-serve key unlocks execution; a small platform fee is included in each built swap, with no subscription.

Can the agent avoid scam tokens on its own?

Yes. The check_token_safety tool rug-gates a token before any swap is built — honeypot detection, mint and freeze authority checks, liquidity-lock and holder-concentration analysis — so the agent can refuse unsafe tokens automatically.

◆ Guide · AI agents & crypto

How to let your AI agent trade crypto — safely, in minutes

Updated 6 August 2026 · Autonomous Intelligence

You can give Claude, Cursor, ChatGPT or your own agent the ability to discover, rug-check and swap crypto — without ever handing over a private key. The trick is a non-custodial flow: the agent builds a transaction, you sign it locally, and only the signed transaction is broadcast. Here is the exact setup, in four steps.

Get a free key → What is this?

01Add the MCP server

Autonomous Intelligence exposes crypto tools over the Model Context Protocol (MCP), so any MCP-compatible AI client can use them. For Claude Code it's one line:

claude mcp add autonomous-intelligence -- npx -y autonomousintelligence-mcp

For Claude Desktop, Cursor, Windsurf or VS Code, paste this into your MCP config and restart the app:

{ "mcpServers": { "autonomous-intelligence": { "command": "npx", "args": ["-y", "autonomousintelligence-mcp"], "env": { "AI_AGENT_KEY": "YOUR_KEY" } } } }

For ChatGPT or any HTTP client, add the hosted server: https://autonomousintelligence.io/mcp. Prefer code? Use the REST rail or a drop-in SDK. Full per-client steps are in the setup guide.

02Get a free key (only for swaps)

Discovery tools — quotes, token safety, gem discovery, token info — need no key at all. To let the agent actually execute a swap, grab a free, self-serve key (no signup, no wallet, no email) and set it as AI_AGENT_KEY:

Generate a key in one click →

03Rug-check before trading

This is the step most integrations skip. Before the agent buys anything, have it call check_token_safety, which flags:

Honeypots (tokens you can buy but not sell)
Active mint & freeze authorities (the deployer can print or freeze)
Unlocked liquidity (the pool can be pulled)
Holder concentration (a few wallets can dump)

A safe agent refuses tokens that fail these checks — automatically, before a swap is ever built.

04Swap non-custodially

The execution flow is deliberately split so the platform can never touch your funds:

get_quote → build_swap (returns an UNSIGNED tx + build_id) → [sign locally with the user's wallet] → submit_swap (broadcast)

build_swap only ever returns an unsigned transaction; submit_swap only ever receives one the agent already signed. Your signing key stays in your environment the whole time. That is what makes it genuinely non-custodial — a flow the platform cannot bypass, not just a promise.

That's it. Your agent can now discover tokens, rug-check them, quote and swap — and you never gave anyone custody of your keys.

FAQ

Do I have to trust the platform with my private key?

No — build_swap returns an unsigned transaction, the agent signs it locally with your own wallet, and submit_swap broadcasts the signed one. The platform never receives your key.

Which AI clients can trade crypto this way?

Any MCP client — Claude Code, Claude Desktop, Cursor, Windsurf, VS Code (Cline/Continue), ChatGPT via the hosted endpoint — plus any custom agent using the REST API or the drop-in SDKs.

How much does it cost?

Discovery, quotes and safety checks are free and keyless. A free self-serve key unlocks execution; a small platform fee is included in each built swap, with no subscription.

Get a free key → See all the agent tools