MCP Native URL security scanner that protects AI agent workflows.

What the MCP-native URL security scanner that protects AI agent workflows. MCP server does

PreClick exposes URL security checks through the Model Context Protocol. It analyzes links for phishing, malware, and other security threats before an agent navigates to them. The service can also compare a URL with the agent's stated purpose, helping identify destinations that do not fit actions such as logging in, making a payment, booking a service, or downloading a file.

The MCP-native URL security scanner that protects AI agent workflows. MCP server is intended for authorized security assessment and agent governance. Suitable uses include validating links before an autonomous browser action, checking URLs embedded in emails or messages, and adding URL checks to automated workflows.

How it works

Clients connect to the hosted Streamable HTTP endpoint at https://preclick.ai/mcp. The default hosted deployment is stateless and accepts JSON-RPC requests through POST /mcp. Clients should send the standard MCP Accept and MCP-Protocol-Version headers. Stateful deployments may require an initialization request and an Mcp-Session-Id on later calls.

A scan can run synchronously by omitting the task parameter, or asynchronously by including task details. Asynchronous requests return a task identifier and processing status. Clients can then use the status and result polling tools to track completion. The intent-aware scan accepts an intent string of up to 248 characters and returns an alignment value such as misaligned, no_mismatch_detected, inconclusive, or not_provided.

When high-confidence analysis confirms certain policy conditions, the response can include a DENY directive. Examples include a confirmed mismatch between the stated intent and destination, or insufficient verification for some high-impact service claims. These policy decisions do not change the risk score.

Setup and configuration

Add the hosted endpoint to an MCP client using the streamable-http transport:

{
  "mcpServers": {
    "preclick-mcp": {
      "transport": "streamable-http",
      "url": "https://preclick.ai/mcp"
    }
  }
}

The trial deployment does not require credentials for up to 100 requests per day. Higher limits and stable quotas require an X-API-Key supplied by CybrLab.ai. An authenticated configuration adds that header to requests. The hosted service is not documented as requiring an API key for the trial tier.

Tools and capabilities

Available MCP tools include:

• url_scanner_scan for standard URL analysis.
• url_scanner_scan_with_intent for analysis with optional task context.
• url_scanner_async_scan and url_scanner_async_scan_with_intent for compatibility-oriented asynchronous submission.
• url_scanner_async_task_status for checking task progress.
• url_scanner_async_task_result for retrieving a completed result.

The MCP-native URL security scanner that protects AI agent workflows. MCP server supports direct URL scanning and task-based processing. Separate Python and Node.js client libraries are also documented, but they are standalone clients rather than MCP server installation methods.

Limitations and notes

Use the service only on systems or websites that you own or are explicitly authorized to assess. The hosted endpoint's default mode is stateless; some clients may probe GET /mcp for an SSE stream, but the hosted deployment returns 405 Method Not Allowed for that request and expects clients to continue with POST /mcp.

Intent analysis is optional and may be inconclusive when evidence is limited. Low-information or instruction-like intent strings are treated as if intent was not supplied. Direct synchronous calls may also be subject to timeout behavior, so asynchronous task submission is the documented recommended mode for task-augmented scans.