Sentrint scans GitHub repos of AI-built apps for security issues and writes a fix prompt. Free plan, $10 scan packs, Founder plan from $22/month.
Researched and fact-checked by Bartu Cavusoglu · Updated November 19, 2026
Sentrint is a security scanner for apps built with AI coding tools. You give it the URL of the GitHub repository behind an app made with Claude, Cursor, Bolt, v0 or Lovable, and it checks the code for problems that AI-written software often carries. These include hardcoded secrets, open database access rules, vulnerable dependencies and unguarded code paths.
The output is meant for people who are not security specialists. You get a score out of 100, a grade, and each finding explained in plain English with its file and line. Sentrint then writes a fix prompt you can paste into the tool that built the app.
The product is labeled Beta (v0.9.3). It reads code only and does not run your app or test the live site, so it is a code scan rather than a penetration test.
A scan has three passes. The first uses fixed rules over your code and lockfiles, with no AI involved. A rule matches or it does not, and this pass alone decides what counts as a finding. Dependency advisories are exact version matches verified without AI.
The second pass is an AI review. It rereads every critical and high finding to reduce false positives. It cannot invent a finding or remove one from your report, but it decides which findings earn a place in the fix prompt and says how many it held back.
The third pass writes the fix. One prompt, ordered worst first, is phrased for the tool that built your app and tells it to change only the flagged lines. All three passes run inside a single-use Google Cloud Run job that is destroyed when the scan ends.
Each report lists every finding in plain English and opens into a detail view with the file, line and risk. An attack path view shows the likely route to a flaw, with each step mapped to a MITRE ATT&CK technique. Sentrint notes this is a likely route and not proof that the flaw is exploitable.
A code map draws your files as a graph sized by risk, so you can see which ones cost your score the most. A README badge shows your live grade, and only grades A to C are displayed. Below C it reads unrated.
For repeat use there are 90 days of searchable audit logs, CSV and JSON downloads, and an MCP server. The MCP server lets an AI agent start a scan, read findings with file and line, and fetch the fix prompt without leaving the chat.
Sentrint is a security scanner for apps built with AI coding tools. It reads the GitHub repo behind an app and reports hardcoded secrets, database access problems, vulnerable dependencies and risky code paths in plain English. It also writes a fix prompt for the tool that built the app.
Sentrint has a free plan with one scan per month. A Top-up pack of 6 scans costs $10 as a one-time purchase. The Founder plan costs $22 per month or $220 per year.
Yes. The free plan includes one scan per month with no card and no trial clock. The site says it is a real scan using the same engine as every paid plan.
It writes fix prompts for 16 tools: Cursor, Windsurf, GitHub Copilot, Cline, Replit, Claude Code, Codex CLI, Gemini CLI, Lovable, Bolt, v0, Base44, Claude, ChatGPT, DeepSeek and Gemini. It also works with any repo on GitHub.
No. Sentrint reads code only. It does not run your app or test the live site, and the company describes it as a code scan, not a pentest.
Each scan runs in a single-use Google Cloud Run job with read-only access to the repo you pick. The clone is deleted when the job ends and is not used to train models. For critical and high findings it keeps about ten lines around the flagged line for 30 days so the fix prompt can be rebuilt.
It is aimed at founders and small teams without a security specialist. The free plan and the $10 pack keep the cost low, and the Founder plan adds multi-repo scanning and exports.
Yes. Its MCP server gives an agent five tools: start a scan, check it, read the findings, get the fix prompt and list past scans. Findings include the file and line.