Give your AI assistant
a scam radar.
SafeBrowz plugs into Claude, ChatGPT and any MCP client, so your assistant checks a link before you click, buy, log in, or sign - phishing, fake logins, crypto wallet-drainers, brand lookalikes and more, in one call.
https://api.safebrowz.com/mcp
Add it as a custom connector - your assistant discovers the tools automatically.
Add it in three steps
Open connector settings
In Claude: Settings → Connectors → Add custom connector. ChatGPT and Cursor have the same "add MCP server" option.
Paste the URL
Drop in https://api.safebrowz.com/mcp and connect.
Activate & ask
Activate with your SafeBrowz Premium key, then just ask "is this link safe?" - or let the assistant check on its own before it acts.
Four tools, one detection engine
Check a URL
One clear verdict on any link - safe, caution, or danger - plus what to do next: proceed, ask you first, or block.
Check many at once
Scan a whole page of search results or every link in an email before your assistant opens any of them.
Find a brand's real domain
Confirm the real site for a brand so your assistant spots the lookalike before you type a password or connect a wallet.
Domain threat history
Ask whether a site has been reported as a scam before - a domain flagged once is remembered.
The scams, in one check
- Phishing & fake login pages that steal your password
- Crypto wallet-drainers, seed-phrase & recovery-phrase theft
- Brand impersonation, typosquats & lookalike domains
- Fake CAPTCHA / ClickFix malware prompts
- Fake online stores & too-good-to-be-true prices
- Tech-support, giveaway & airdrop scams
- Clipboard hijack (pastejacking) & hidden overlays
- Disposable domains & redirect cloaking
Three layers, not a blocklist
Local rules
URL patterns, typosquat & homoglyph checks, disposable-TLD rules and a 550+ brand database - instant, before anything is fetched.
Trusted feeds
Google Safe Browsing, PhishTank and URLhaus, plus domain-age / WHOIS and shared threat intelligence.
AI content analysis
For anything the first two cannot settle, an AI model reads the actual page in 100+ languages to catch brand impersonation and social engineering.
Simple, and separate from your other SafeBrowz apps
- 1,000 checks / month
- All four tools
- Reads pages in 100+ languages
- A separate key, just for your AI assistant
- Pay by card or USDC
- For agent fleets & teams
- Higher monthly volume
- Bearer API key, usage invoicing
- Same one detection engine
Questions
What is the SafeBrowz MCP connector?
It plugs SafeBrowz's scam & phishing detection into your AI assistant (Claude, ChatGPT, Cursor, or any Model Context Protocol client), so the assistant can check any link before you click, buy, log in, or sign.
Which AI assistants does it work with?
Any MCP client. Today that includes Claude connectors (claude.ai, Claude Desktop, Cowork and Claude Code), ChatGPT connectors, Cursor, Cloudflare OS via its MCP Server Portal, and agent frameworks. You add it as a custom connector with the URL https://api.safebrowz.com/mcp.
Does it affect my SafeBrowz browser extension or app?
No. The connector uses its own separate SafeBrowz key, so it never touches your extension or app usage. Each SafeBrowz key works on one surface, so your AI assistant gets its own dedicated key.
Is it secure?
Yes. Activation uses OAuth 2.1 with PKCE, tokens are hashed, and payment happens on the processor's secure page - SafeBrowz never sees your card. No per-user browsing history is stored.
How much does it cost?
SafeBrowz Premium is $14.99 per year and includes 1,000 checks per month. Pay by card or USDC. Higher volume and agent fleets use an enterprise key.
What's the difference from the browser extension?
The extension protects the pages you open in your browser. The connector protects the actions your AI assistant takes on your behalf - checking links it is about to visit, recommend, or act on.
How do I use SafeBrowz in a Cloudflare Worker or agent?
A Worker connects to the connector as an MCP client. Store your key as a Worker secret and send it as an Authorization: Bearer header. Use the Cloudflare Agents SDK or the Vercel AI SDK MCP client so the model calls check_url on its own, or POST a JSON-RPC tools/call to https://api.safebrowz.com/mcp directly. For plain URL checks without an agent tool, the REST API at https://api.safebrowz.com/v1/detect is simpler.
Ready in about a minute
Copy the URL, add it as a connector, activate. Your assistant handles the rest.
https://api.safebrowz.com/mcp