Give Claude x-ray vision on any site.
Boolsai MCP turns the stack-intelligence scanner into a tool Claude can call. Ask your agent "what is gymshark.com running?" — it scans the site live and answers with the full vendor list, every pixel, every API the page calls, and the internal route map. One JSON response, ~5 seconds.
Type a domain. See what the AI sees.
This widget hits the real https://intel.boolsai.ai endpoint from your browser. Same call your agent would make. Same JSON.
View raw JSON response
curl 'https://intel.boolsai.ai/?url=stripe.com'
Three ways to plug Boolsai into Claude.
No keys, no signup. The endpoint is public and rate-limited per IP. Pick the path that matches how you use Claude.
Paste one URL into Claude. Done.
In Claude.ai, open Settings → Connectors → Add custom connector and paste the URL. Every chat then has a boolsai_scan tool Claude can call.
https://boolsai.ai/mcpOne block into your Claude Desktop config.
Paste into claude_desktop_config.json, restart Claude. The Boolsai tool shows up in the tool picker — no key, nothing to sign up for.
{
"mcpServers": {
"boolsai": {
"command": "npx",
"args": ["-y", "@boolsai/mcp"]
}
}
}Building your own agent? Drop the tool in.
Add the definition to your tools[] array. When Claude calls it, fetch https://intel.boolsai.ai/?url=… and hand the JSON back as the tool_result. No auth.
{
"name": "boolsai_scan",
"description": "Scan a public website. Returns vendors, hosts, endpoints, and routes as JSON.",
"input_schema": {
"type": "object",
"properties": {
"url": { "type": "string" }
},
"required": ["url"]
}
}Not a list of guesses. A map of what's actually running.
Every external host
Each domain the page talks to, grouped by category — analytics, pixel, CDP, reviews, search, support, personalisation, consent, CDN. Including the ones loaded by other scripts at runtime.
Every endpoint fired
Fetch, XHR, beacon, and inline-script calls — captured as the page actually behaves in a browser. Method, URL, and trigger context. The stuff passive directories never see.
The structural DOM tree
Internal route trie, head metadata, hreflang map, inline scripts, JSON-LD payloads, and asset CDNs. Enough signal for Claude to answer questions about the site, not just list it.
The web is JavaScript now. Your agent reads HTML.
When Claude opens a modern site with a fetch, it gets a shell — most of the stack loads after the page renders. Boolsai runs the page in a real browser, captures the runtime, and hands Claude a JSON map of the whole thing. That's the difference between an agent that can read a site and one that can reason about it.
<script src> tagsWhat agencies are building on the Boolsai MCP.
"Prep me for tomorrow's pitch"
Drop a prospect URL into Claude. It scans, summarises the stack, finds gaps, and writes the first page of the audit deck — before you've poured your coffee.
"Watch my client book"
A scheduled agent scans every client domain weekly, diffs against the last scan, and DMs the account lead the moment tracking breaks or a vendor swaps.
"Watch the market"
A competitor watchlist with Claude on the trigger. When a brand drops Klaviyo for Iterable or adds a new pixel, your agent writes the brief and pushes it to Slack.
Plug it in. No key, no signup.
The endpoint is public and rate-limited per IP. If you need higher limits or a private mirror for client work, email founder@boolsai.ai.