AgentFuelblog

Give your AI agent a rug-check in two minutes

If your agent can trade on Solana, it can also lose money on a token that was built to be dumped. Here is how to let it ask for a risk verdict first, using MCP. No account, no API key, no wallet needed for the free trial.

Step 1. Add the MCP server

AgentFuel runs an MCP server at https://api.agentfuel.eu/mcp (Streamable HTTP). In Claude Code:

claude mcp add --transport http agentfuel https://api.agentfuel.eu/mcp

In Claude Desktop, Cursor or another client that speaks stdio, use mcp-remote in your config:

{
  "mcpServers": {
    "agentfuel": { "command": "npx", "args": ["-y", "mcp-remote", "https://api.agentfuel.eu/mcp"] }
  }
}

The server is also listed in the official MCP registry as io.github.Fizzl13/agentfuel.

Step 2. Ask before you buy

Your agent now has these tools, among others:

ToolUse it when
rugcheckYour agent is about to buy a Solana token. Returns a 0 to 100 risk score with reasons.
latest_launchesYour agent looks for new pump.fun tokens. Filter on maxRisk so it only sees launches without red flags.
safe_buyYour agent has built the buy transaction. One green, orange or red verdict on the token and the transaction together.
injection_checkYour agent reads a token description or a message from a stranger. Checks whether the text tries to hijack it.

Try a prompt like this:

Before you buy any Solana token, call the AgentFuel rugcheck tool on its mint.
Do not buy when the level is "high". Tell me which findings you saw.

Put that rule in your agent's system prompt and it becomes a habit, not a one-off.

Step 3. Read the answer

The findings list is where the value is. A few codes your agent should treat as a hard stop:

Pool accounts and pump.fun's Mayhem Mode agent are left out of the holder math automatically. How the score works in detail: How we compute a rug score.

Step 4. When the free trial is not enough

MCP gives 50 free tool calls per IP address per day. A bot that checks every launch needs more. Then switch to the HTTP endpoints and pay per call with x402: 0.002 USDC per rug-check, 0.001 USDC per launch-feed call, 0.01 USDC per safe_buy. Your bot pays from its own wallet; there is no subscription and nothing to sign up for. See Paying with x402.

The score is an automated on-chain check, not financial advice. A low score means no red flags were found, not that a token is safe.
Next stepRead how the score is built, or start with the docs: How we compute a rug score · Docs