How does my AI agent verify URLs it receives from other AI agents?

The problem

Autonomous AI agents receive URLs from other AI agents, from user input, and from web scraping. Before following a link, rendering a page, or passing a URL along, an agent needs to decide whether to trust it. Human-readable warnings don't help here. Agents need structured data they can apply their own thresholds to.

Without verification, one compromised URL in a multi-agent pipeline can propagate through the entire system.

How Unphurl solves it

Unphurl returns structured, machine-readable signals. No opinions, no verdicts. Your AI agent gets raw data across 7 dimensions and applies its own logic. A score, a breakdown of what fired and why, and the full signal payload for any custom decision-making.

Signals that matter for this use case

• – Domain age <7 days flags newly created domains that may be attack infrastructure
• – Domain age <30 days identifies recently staged infrastructure worth extra caution
• – SSL invalid signals compromised or abandoned sites
• – Chain incomplete means the URL destination couldn't be reached
• – No MX record indicates a domain without real email infrastructure
• – URL structural analysis catches IP addresses, encoded hostnames, and high-entropy domains

Suggested scoring profile

{
  "name": "agent-trust",
  "weights": {
    "brand_impersonation": 40,
    "domain_age_3": 35,
    "domain_age_7": 25,
    "domain_age_30": 15,
    "ssl_invalid": 15,
    "chain_incomplete": 20,
    "no_mx_record": 10,
    "url_contains_ip": 15,
    "domain_entropy_high": 10,
    "encoded_hostname": 10,
    "redirects_5": 20,
    "phishing_floor": 80
  }
}

What a result looks like

Another AI agent passes your agent a URL to process: https://data-api.xyz/export/results. Your agent checks it before following:

Score 35 is below the agent's threshold of 50. The agent proceeds but logs the signals for audit. If the score were higher, the agent would reject the URL and request an alternative.

Cost

AI agents typically work with a mix of well-known APIs (cached or Tranco, free) and unknown endpoints. An agent processing 500 URLs per day might see 50-150 pipeline checks. The Standard package ($39 for 500 pipeline checks) covers most agent workloads for weeks.

Get started

The MCP server is the natural fit for agent-to-agent trust. Install it in your AI tool and use the companion skill for proactive checking:

{
  "mcpServers": {
    "unphurl": {
      "command": "npx",
      "args": ["-y", "@unphurl/mcp-server"],
      "env": {
        "UNPHURL_API_KEY": "uph_your_key_here"
      }
    }
  }
}

With the companion skill installed, your AI agent checks unfamiliar URLs automatically before following or recommending them.