Kipra Scout · Free

See the AI on any website.

A free Chrome extension that shows what a stranger can make your AI chatbot do, and how to fix it.

  • Chat endpoints anyone can resend, and what that costs you at list prices
  • The tools your AI can be talked into calling, and the data anyone can reach
  • AI keys leaked in page code, and LLM APIs called straight from the browser
  • Vibe-coded backends with open tables, with the fix written for your site
  • A Profile tab with the site's public setup: DNS, hosting, headers, third parties, email

Scout reads what the page already loaded. Nothing about you or the page is sent to Kipra.

Get Scout

Leave your email and the download starts right away.

Chrome, Edge, Brave and other Chromium browsers. Version 1.6.0.

What Scout shows you.

At most five items to review, in order, each with fix steps for your site and ticket text to paste into GitHub, Linear or Jira.

Scout recording a chatbot: the chat API takes model settings from the browser with no auth, about $0.51 per 1,000 requests a stranger sends, with fixes for the endpoint
Record your own chatbot. Use your chat while Scout watches. It shows who can resend the call and what that costs you.
Scout showing an open Supabase table with Row Level Security SQL on the owner column and a check to run
Open data, with the fix. Row Level Security SQL for the real table and owner column, plus a check to run.
Scout showing the tools a chatbot can call, the knowledge base anyone can copy, and which personal-data fields go to which AI vendor
What your AI can reach. Tools ranked by harm, and which personal-data fields go to which AI vendor. Names only, never values.
Scout showing what changed since the last visit, with the fix for a new item
What changed since last time. New items and ones no longer seen, kept on your device only.
Scout showing known-exploited CVEs and end-of-life server software
Exploited CVEs and old software. Checked offline against CISA's exploited list. No version is sent anywhere.

What it finds.

Scout watches what the page actually does and reports an observation only when the evidence supports it. Each one is marked proven or to confirm.

A web page with what Scout found around it, from a leaked key to old software, with one open chat endpoint marked as proven sk-proj-…Xa9f system prompt orders · anon mcp /sse jquery 1.12.4 POST /api/chat Proven · no auth

Critical

AI provider keys (OpenAI, Anthropic, Groq, Hugging Face, xAI…), Supabase service keys, Stripe secret keys, database URLs with a password, .env or .git served live.

High

A provider API called from the browser, a chat endpoint where the browser picks the model or system prompt with no credentials, MCP or Ollama answering anonymously, Supabase tables written with the public key.

Medium

Hidden instructions planted for AI agents, a system prompt in page code, exploited CVEs that may apply, end-of-life server software.

Low

Chat widgets, in-browser models, published source maps, missing security headers.

Private by design.

Always on

Reads what the page already loaded: scripts, DOM and the shapes of its requests (field names, never content). Sends nothing to Kipra or anyone else.

Profile tab

When you open it, at most once a day per site, Scout looks up the site's public records by host name. Never the page URL, cookies or anything about you.

Active scan

Only after you click and tick that you own the site or may test it: about 61 read-only GETs to the site's own origin, no cookies.

No accounts, analytics, telemetry or remote code. Results are gone when the tab closes. Full details in the privacy policy.