MCP server + REST API

The safety guardrail MCP server for Claude and autonomous agents.

An agent that browses, reads email, or follows a link out of a tool result can be steered by a malicious URL exactly like a person can. SkanQRCode is a URL safety check built for the agent tool-call path: one call, a verdict, before the fetch.

Request early access Building a QR scanner instead?

check_url tool response

{
  "verdict": "malicious",
  "action": "block",
  "mode": "url",
  "reasons": ["ACTIVE_THREAT_FEED_MATCH", "KNOWN_MALWARE_MATCH"],
  "cached": false,
  "executionTimeMs": 38,
  "environment": "production",
  "licensedForProduction": true,
  "requestId": "req_01J9Z8QAENP0S2"
}

Quickstart

One command. Then wire it into Claude Desktop.

1. install

npx -y @skanqrcode/mcp-server

# claude_desktop_config.json

{
  "mcpServers": {
    "skanqrcode": {
      "command": "npx",
      "args": ["-y", "@skanqrcode/mcp-server"],
      "env": {
        "SKANQRCODE_API_KEY": "sk_test_..."
      }
    }
  }
}

Live on npm as @skanqrcode/mcp-server. Get a free sk_test_ key (1,000 checks/mo, no card) at app.skanqrcode.com.

MCP tool schema

check_url

Evaluates whether a URL is safe to fetch or scrape. Returns an allow, warn, or block verdict in real time.

# wherever your agent decides to fetch a URL:

var req = URLRequest(url: URL(string: "https://api.skanqrcode.com/v1/check")!)
req.httpMethod = "POST"
req.setValue("Bearer \(apiKey)", forHTTPHeaderField: "Authorization")
req.httpBody = try JSONEncoder().encode(["target": url])
let (data, _) = try await URLSession.shared.data(for: req)
let result = try JSONDecoder().decode(CheckResponse.self, from: data)
if result.action == "block" { return } // never UIApplication.shared.open(url)

Why agents need this too

Same phishing link. New way to click it.

Zero-day, not just feeds

Domain age, brand-lookalike scoring, and shortener unwinding run on every check, so the URL nobody has reported yet still gets caught.

MCP or a direct POST

A check_url tool for Claude Desktop and other MCP clients, or the same endpoint for LangChain, LlamaIndex, or a custom agent loop.

Verdicts change, you get told

If a URL your agent already fetched gets raised to suspicious or malicious later, a webhook tells you within minutes, on Pro and Business.

Pricing

Free to wire it up. $19.90 when the app is live.

Sandbox action: block and the MCP server are free, so you can verify the integration before you pay for it. A production license, higher limits, and verdict-change webhooks live on Pro.

Free

$0/mo

1,000 API calls / mo

Wire it up. Test every verdict, sandboxed.

Fast path + sample JSON
action: block in sandbox mode
licensed_for_production: false
10 rpm
Agent API + MCP (1,000 req/mo)
No verdict-change webhook

Join waitlist

Pro

$19.90/mo

50,000 API calls / mo

The app is live. You can stop the open.

action: block in production
Lookalike + NRD + live reputation checks
30 rpm
Agent API + MCP, production licensed
Verdict-change webhook
Customer allowlists

Request Pro access

Flexible overages at $0.001 per additional check, so production never throttles.

Business

$169.90/mo

500,000 API calls / mo

Volume is the product.

Everything in Pro
60 rpm + SLA 99.9%
Deep path included
DPA
Custom volume available

Talk to us

Early access

Keys when curl returns JSON. Not a newsletter.

Traction for us is a package name and a request log. If your agent fetches whatever a page or tool result hands it, we want that call.