Agent Input Guard — Prompt Injection & Agent Security

About this service

Scan untrusted text before it enters an LLM context or tool chain. Detects prompt injection, instruction override, credential exfiltration, tool abuse, encoded-instruction hints and data-exfiltration signals. Returns risk, score, findings and a recommendation. Deterministic English-language heuristics, not a guarantee of safety; false positives and missed attacks are possible. Send {"text":"untrusted content"}; maximum 100,000 characters. Do not send secrets.

Tools included — 1

Reported by this server's own tools/list, refreshed automatically.

Add to your agent

This listing is paid, so it is not a server your client connects to — the seller's endpoint stays private and the call is settled through the FiatDock gateway. Install FiatDock once, then invoke this listing by id.

1. Paste this into your MCP client config (Claude Desktop, Cursor, Windsurf, Gemini CLI). Click the block to select all, then copy.

{
  "mcpServers": {
    "fiatdock": {
      "command": "npx",
      "args": [
        "-y",
        "fiatdock-mcp"
      ],
      "env": {
        "AGENT_PRIVATE_KEY": "0x..."
      }
    }
  }
}

Replace 0x... with the private key of a Base wallet that actually holds USDC — that key is what signs the payment. Leave the placeholder in and nothing crashes, but the call below returns a 402 price quote instead of buying.

2. Then have your agent call:

call_service({ id: "svc_26b82fa0-f9ae-4e33-aad6-b4e10cc2c6c2", args: { … } })

Listing id: svc_26b82fa0-f9ae-4e33-aad6-b4e10cc2c6c2 · networks: base · paid per call via x402 (100% to the seller — 0% launch offer, the buyer makes a single x402 payment)
Not sure what args to send? get_service({ id: "svc_26b82fa0-f9ae-4e33-aad6-b4e10cc2c6c2", includeSchemas: true }) returns the argument names and types — free, and it is the only way to get them.

Reviews — none yet

Only buyers who completed a paid call can review — verified purchases only.