PreClick MCP Server

> PreClick — An MCP-native URL preflight scanning service for autonomous agents. It scans links for threats and confirms they match the intended task before execution. Built for agentic workflows, it provides high-accuracy, context-aware browsing governance with adaptive learning.

Publisher: CybrLab.ai | Service: PreClick

Hosted Trial Tier: No API key required for up to 100 requests/day. For higher limits and stable quotas, use an API key (contact contact@cybrlab.ai).

Overview

PreClick is a URL security scanner and MCP server that enables AI agents and any client to analyze URLs for phishing, malware, and other security threats before navigation. Use it as a standalone URL scanner via the client libraries below, or connect directly via the MCP protocol.

Integrations

Client Libraries (Recommended)

The fastest way to add URL security scanning to any AI agent, automation pipeline, or application. These standalone URL scanner clients handle connection, polling, and error recovery out of the box — no MCP protocol knowledge or conformance required:

Both libraries provide a simple scan-oriented URL security scanning API (`scan(url)` / `scanWithIntent(url, intent)`) that returns results directly — phishing detection, threat analysis, and intent alignment in a single call. No protocol vocabulary, no `connect()` boilerplate, no manual polling for the common case.

Framework Adapters

Manual MCP Configuration

For MCP-native clients that speak the protocol directly, see Quick Start below.

Authentication Modes

Important Notice

This tool is intended for authorized security assessment only. Use it solely on systems or websites that you own or for which you have got explicit permission to assess. Any unauthorized, unlawful, or malicious use is strictly prohibited. You are responsible for ensuring compliance with all applicable laws, regulations, and contractual obligations.

Use Cases

• Pre-flight URL validation for AI agents
• Automated URL security scanning in workflows
• Malicious link detection in emails/messages

Quick Start

1. Configure Your MCP Client

Choose one option:

Trial (hosted, up to 100 requests/day without API key):

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

Authenticated (recommended for stable and higher-volume usage):

{
  "mcpServers": {
    "preclick-mcp": {
      "transport": "streamable-http",
      "url": "https://preclick.ai/mcp",
      "headers": {
        "X-API-Key": "YOUR_API_KEY"
      }
    }
  }
}

2. Optional: Initialize Session (stateful mode only)

Default hosted usage is stateless. Clients send JSON-RPC messages with `POST /mcp`.

Some Streamable HTTP clients may also probe `GET /mcp` for an SSE stream. On the stateless hosted deployment, `/mcp` does not offer an SSE stream and returns HTTP `405 Method Not Allowed`. Clients should treat `405` as "no SSE stream on this endpoint" and continue using `POST /mcp`.

Clients should still send the standard MCP HTTP headers:

• `Accept: application/json, text/event-stream` on POST
• `MCP-Protocol-Version` on all non-initialize requests

The hosted deployment currently normalizes missing or incomplete POST `Accept` headers for compatibility. It also allows missing `MCP-Protocol-Version` on discovery-only POST list requests (`tools/list`, `resources/list`, `prompts/list`) for registry compatibility. Clients should not rely on either behavior.

# Only required if the server is running in stateful mode
curl -X POST https://preclick.ai/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "X-API-Key: YOUR_API_KEY" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "initialize",
    "params": {
      "protocolVersion": "2025-06-18",
      "capabilities": {},
      "clientInfo": {"name": "my-client", "version": "1.0"}
    }
  }'
# Response includes Mcp-Session-Id header - save it for subsequent requests

3. Start a Scan

`url_scanner_scan` supports two execution modes (the same modes apply to `url_scanner_scan_with_intent`):

• Task-augmented (recommended): Include the `task` parameter for async execution
• Direct: Omit the `task` parameter for synchronous execution

curl -X POST https://preclick.ai/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "MCP-Protocol-Version: 2025-06-18" \
  -H "X-API-Key: YOUR_API_KEY" \
  -d '{
    "jsonrpc": "2.0",
    "id": 2,
    "method": "tools/call",
    "params": {
      "name": "url_scanner_scan",
      "arguments": {
        "url": "https://example.com"
      },
      "task": {
        "ttl": 720000
      }
    }
  }'
# If stateful mode is enabled, include: -H "Mcp-Session-Id: YOUR_SESSION_ID"

Response (task submitted):

{
  "jsonrpc": "2.0",
  "id": 2,
  "result": {
    "task": {
      "taskId": "550e8400-e29b-41d4-a716-446655440000",
      "status": "working",
      "statusMessage": "Queued for processing",
      "createdAt": "2026-01-18T12:00:00Z",
      "lastUpdatedAt": "2026-01-18T12:00:00Z",
      "ttl": 720000,
      "pollInterval": 2000
    }
  }
}

Optional: Provide an url visiting intent for additional context (recommended but not required):

curl -X POST https://preclick.ai/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "MCP-Protocol-Version: 2025-06-18" \
  -H "X-API-Key: YOUR_API_KEY" \
  -d '{
    "jsonrpc": "2.0",
    "id": 3,
    "method": "tools/call",
    "params": {
      "name": "url_scanner_scan_with_intent",
      "arguments": {
        "url": "https://example.com",
        "intent": "Book a hotel room"
      },
      "task": {
        "ttl": 720000
      }
    }
  }'

Recommendation: Use `url_scanner_scan_with_intent` when you can state your purpose (login, purchase, booking, payments, file download) so intent/content mismatch can be considered as an additional signal. Otherwise use `url_scanner_scan`.

Max intent length: 248 characters.

Low-information or instruction-like intent strings are treated as not provided.

Result includes `intent_alignment` (`misaligned`, `no_mismatch_detected`, `inconclusive`, or `not_provided`).

`no_mismatch_detected` is only returned when intent analysis had sufficient evidence; if intent analysis is unavailable or evidence is limited, result is `inconclusive`.

When `intent_alignment` is `misaligned` and confirmed by successful high-confidence analysis, the response directive is `DENY` with reason `intent_inconsistent_destination` (policy gate; risk score is unchanged).

When high-confidence analysis confirms an unverified high-impact service claim with weak identity corroboration in a low-confidence context, the response directive is also `DENY` with reason `insufficient_service_verification` (policy gate; risk score is unchanged).

In additional contextual low-evidence policy cases, responses may return `DENY` with reasons such as `insufficient_service_verification` or `insufficient_trust_signals` (policy gate; risk score is unchanged).

Direct-call timeout note: synchronous tool calls use a bounded server wait window sized for direct-only clients (hosted default 90s). If timeout is reached, the server returns JSON-RPC `-32603` with `error.data.taskId` and `error.data.pollInterval` so you can continue via `tasks/get` / `tasks/result`.

Compatibility note: if your MCP client cannot call native Tasks methods (`tasks/get` / `tasks/result`), use `url_scanner_async_scan` or `url_scanner_async_scan_with_intent` to submit work and then poll with `url_scanner_async_task_status` / `url_scanner_async_task_result`. Call these compatibility tools as ordinary tools only; do not include a native MCP `task` parameter.

4. Poll for Results

`tasks/result` uses a shorter hosted blocking wait (default 30s). If this wait is exceeded, the server returns JSON-RPC `-32603` with `error.data.taskId` and `error.data.pollInterval`. Native Tasks clients should prefer polling with `tasks/get` until status is `completed`, then call `tasks/result` to retrieve the final result immediately.

curl -X POST https://preclick.ai/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "MCP-Protocol-Version: 2025-06-18" \
  -H "X-API-Key: YOUR_API_KEY" \
  -d '{
    "jsonrpc": "2.0",
    "id": 3,
    "method": "tasks/result",
    "params": {
      "taskId": "550e8400-e29b-41d4-a716-446655440000"
    }
  }'
# If stateful mode is enabled, include: -H "Mcp-Session-Id: YOUR_SESSION_ID"

Response (completed task — CallToolResult shape, same as synchronous `tools/call`):

{
  "jsonrpc": "2.0",
  "id": 3,
  "result": {
    "content": [
      {
        "type": "text",
        "text": "{\"risk_score\":0.05,\"confidence\":0.95,\"analysis_complete\":true,\"agent_access_directive\":\"ALLOW\",\"agent_access_reason\":\"no_immediate_risk_detected\",\"intent_alignment\":\"not_provided\"}"
      }
    ],
    "isError": false
  }
}

Available Tools

See Full API Documentation for detailed schemas and examples.

Authentication

Authentication requirements depend on deployment mode:

• Hosted endpoint (`https://preclick.ai/mcp`): API key is optional for up to 100 requests/day.
• Hosted endpoint above trial quota: API key required.

See Authentication Guide for details on getting API keys.

Technical Specifications

Support

• Publisher: CybrLab.ai
• Service: PreClick
• Email: contact@cybrlab.ai

License

Apache License 2.0 - See LICENSE for details.

Copyright CybrLab.ai