Ship AI features
without losing sleep

Sentrint scans the GitHub repo behind an app you built with Claude, Cursor, Bolt, v0 or any of the 16 AI coding tools it supports. It checks the code for hardcoded secrets, database access rules, vulnerable dependencies and risky code paths. An AI review then rechecks a capped number of findings against your code to cut false positives. You get a score out of 100, each finding in plain English, and a fix prompt written for the tool that built the app. Apply the fix and scan again to see what changed. It reads code only: it does not run your app or test the live site.

AI-written code can pass every test and still leak secrets, leave your database open or skip login checks. Sentrint finds them before anyone else does, fix included.

Or see a live demo first
The same repo, before and after its fixes.
Got questions? Ask Gourab Gourab Dasgupta, who answers Sentrint support
Runs on Google Cloud Google for Startups Cloud Program
Single-useInstance DeletedYour clone Read-onlyRepo access
Read the commitments
Fix prompts written for
  • Cursor
  • Windsurf
  • GitHub Copilot
  • Cline
  • Replit
  • Claude Code
  • Lovable
  • Bolt
  • v0
  • Base44
  • DeepSeek
  • Claude

Names and logos belong to their owners. Sentrint is not affiliated with them.

Why this matters

AI coding tools write code that runs. Sentrint reads it for what it exposes: leaked keys, open database rules, unguarded routes, and the fix for each.

98%

of 1,072 Supabase-backed apps built with Lovable, v0, Bolt, Replit and Windsurf had at least one security issue. 26 had none.

Symbiotic Security
11%

of 20,052 indie launch URLs exposed Supabase credentials in frontend code. The anon key is meant to be public. A service_role key never is.

SupaExplorer
1.5M

API tokens and 35,000 email addresses exposed by Moltbook, a vibe-coded app, days after launch. Its Supabase tables had no RLS policies.

Wiz Research

What our users are saying

Sentrint helped me detect several key security vulnerabilities in a vibe coded personal agent I had built for myself. The agent’s vulnerabilities had the potential to leak a lot of private and sensitive data, which were quickly resolved. I’m glad I found this product!!

4.7 / 5
Naman
Google, Software engineer

I am a non technical guy doing a project that has the potential to become a good product. I have little knowledge in cybersecurity and don’t have the funds to get a professional. I had mistakenly thought Claude can do things for me. I know better now thanks to a costly mistake. Apparently I have to know what issues to fix and then instruct AI to do it. This is where Sentrint became my go to. It has a lot of rules baked into the product that can check these issues for me. I’m satisfied and I would recommend Sentrint to all non tech people out there who is new to product building.

4.5 / 5
Verified Sentrint user Verified
UX Specialist

Easy to find code vulnerabilities, detect where code lacks and need further improvements.

4.5 / 5
Verified Sentrint user Verified
Data Scientist

What one scan gives you

01 · Risks

Every security risk across your repos, worst first

37.6%

more critical vulnerabilities after five rounds of AI edits, in a controlled study of 400 code samples. Shukla et al., 2025

Scan again and it tells you what you fixed and what is new.

The Findings page across a workspace: 2,099 open risks, 1,074 critical or high, 570 with a fix ready, a weekly new versus fixed chart, and a grid of code, dependency and secret risks by severity.
Findings · every repo, one list

02 · All findings

Each vulnerability explained in plain English

44%

of tested AI code-generation tasks produced code with a known vulnerability. Veracode, 2026

Each row says what the AI review made of it: confirmed, unclear or a false positive.

The all findings table: filters for status, severity, age and repo, then critical rows across several repos, each with a plain English explanation, its file and repo, and an AI verdict tag reading AI-confirmed or unclear.
All findings · code, worst first

03 · Finding detail

Open a finding to see the file, the line and the risk

88%

of 10,616 publicly leaked AWS keys still worked when researchers tested them again. Truffle Security, 2026

Leaked keys get tested, so you know which one to rotate first.

A finding opened in a side panel: a critical SQL injection, user input concatenated directly into a SQL query, with its status, repo and commit, the date found, the file and line, and an AI-confirmed verdict.
Finding detail · real scan of a public practice repo

04 · Attack path

How an attacker could reach the flaw

T1190

Exploit Public-Facing Application. Every step names its MITRE ATT&CK technique.

It shows the likely route, not proof that this one is exploitable.

The attack path for a SQL injection: step one, initial access, the file handles incoming requests and is reachable from a request, ATT&CK T1190; step two, the weakness itself; then the AI review's reasoning and what it assumed.
Attack path · mapped to MITRE ATT&CK

05 · Fix prompt

A fix prompt for the AI tool that wrote the code

16

AI coding tools, each with a prompt phrased for it.

Ordered by what is exploitable now, and told to change nothing else.

A fix prompt panel with a button per AI tool and a prompt preview: fix 25 critical vulnerabilities first, 9 of them leaked credentials that only a human can rotate.
Fix prompt · copy, paste, rescan

06 · Code map

A map of your code that shows which files to fix first

94

files drawn as one graph, sized by risk, in a real scan of a public practice repo.

Files are ranked by what they cost your score.

A code map of a deliberately vulnerable public repo: 94 files in 21 folders drawn as a graph, risky files as larger orange nodes with critical and high findings as diamonds, beside a ranked list of files to fix first.
Code map · real scan of a public practice repo

07 · Audit logs

A log of every scan, sign-in and change

90 days

of sign-ins, scans, status changes and API keys, kept and searchable.

Every view downloads as CSV or JSON, and the download is logged too.

The audit log filtered to scan events: filters with counts on the left, a bar timeline of activity, then timed rows such as an issue status changed from open to ignored, a fix prompt built and the scan workspace deleted.
Audit logs · filtered to scan events

08 · Badge

A README badge that shows your current grade

A to C

are the only grades a badge shows. Below C it reads unrated.

It reads the live grade every time the README loads.

09 · MCP

Scan from your AI agent, without leaving the chat

5

tools your agent can call: start a scan, check it, read the findings, get the fix prompt, list past scans.

Findings come back with the file and line, so the agent can fix them in place.

A Claude Code session connected to the Sentrint MCP server. Asked to check a repo for critical issues, it calls sentrint_list_scans, which returns a finished scan with score 4, then sentrint_get_findings for critical findings, which returns 48. It then answers with what to fix first: command injection and debug mode at named files and lines, a dependency upgrade in requirements.txt that clears about 11 CVEs, and two SQL injections.
MCP · a real Claude Code session on a public practice repo

Three passes over your code, only the first one decides

  1. 01 No AI

    The engine finds it

    • Fixed rules, over your code and your lockfiles
    • A rule matches or it does not
    • Nothing here can be talked into a finding
  2. 02 AI

    A second read cuts the noise

    • Every critical and high finding, read a second time
    • Cannot invent one. Cannot take one off your report
    • Decides what earns a place in the fix prompt, and says how many it held back
  3. 03 AI

    The fix gets written

    • One prompt, ordered worst first
    • Written for whichever tool built your app
    • Carries its own limits: change these lines, keep every signature
Where it runs

All three passes happen inside one single-use Google Cloud Run job, created for your scan and destroyed when it ends.

What happens to your code

  • Built to DPDP Act 2023

    India's data-protection law, and the one we answer to. Consent you can withdraw, a breach clock, and a named officer on the hook.

    How we meet it
  • Aligned with GDPR data rights

    Your access, correction and deletion rights, honoured wherever you live. We claim no certificate and no EU representative, only the substance.

    What aligned means
  • Payments PCI DSS via Dodo

    Card entry is handled entirely by our payment processor, Dodo. Those details never reach, and are never stored on, a Sentrint server.

    Where money goes

Subprocessor directory

  • Cloudflare DNS, CDN and the firewall in front
  • GitHub Sign-in, and reading the repo you pick
  • Upstash Queues the scan, holds the rate limits
  • Google Cloud Runs the app, the scan job and our mail
  • OpenRouter Carries the snippet to the model
  • Anthropic Writes the fix, zero-retention endpoints
  • Supabase Holds your account and your findings
  • Dodo Payments Takes payment as merchant of record
  • Sentry Crash reports, with identifiers off
  • PostHog Which pages get used, no email or IP

What it costs

Top-up

Buy a pack, scan when you need to. No subscription and no renewal date.

₹799$10 one time

6 scans in a pack.

Buy 6 scans
  • 6 deep scans, credits never expire
  • Every finding, ranked by what to fix first
  • Copy-paste fix prompt for 16 AI tools
  • Per-finding AI fix & false-positive review
  • Score, grade and a badge for your repo
  • One-time purchase, not a subscription
Most popular

Founder

For anyone shipping every week. Scan on every change instead of rationing them.

Yearly 2 months free
₹1,799$22 a month ₹17,990$220 a year

Cancel any time, from Settings.

₹1,499.17$18.33 a month, billed once a year.

Subscribe Subscribe
  • 36 scans a month, everything in Top-up
  • Top-up packs when you need more scans
  • CSV & JSON export of findings
  • Dependency inventory & CycloneDX SBOM
  • MCP server
  • REST API

Services

External exposure intelligence for people and organizations. Hands-on work, not automation or software. Every project is scoped on a call before a quote is provided.

from ₹12,000$150

₹4,000$50 a day, 3-day minimum.

Coming soon

  • Digital Footprint Assessment
  • Breach & Credential Exposure Assessment

Free plan One scan every calendar month, not every thirty days. Your GitHub account needs to be 30 days old to claim it. 1 scan per month

  • Score, grade and badge
  • Every finding, in full
  • The fix prompt
Free scan ₹0$0

Compare everything

Find out what your app is leaking

Listed with