Made for apps built with Cursor
Scan what Cursor shipped, not the chat
Cursor can see the repo. It cannot see the bundle the browser downloads after the build. Sekrd scans that URL: client env secrets, a vulnerable React Server Components pin, middleware that the bypass header opens, source maps, and an MCP server that lists tools with no credentials.
Free scan. Pre-Launch Audit is $39 once. Continuous Pro is $24/month billed annually ($288/year), or $29/month billed monthly.
Risks Sekrd checks on a Cursor-built app
Free scan
A client env prefix on a server secret
NEXT_PUBLIC_, VITE_, EXPO_PUBLIC_, NUXT_PUBLIC_, and PUBLIC_ inline the value into the browser. The free scan reports that prefix when the variable name is a server secret: model API keys, a Stripe secret, service role, a database URL, a webhook secret, and the other names in the jssecurity rule. Public-by-design names such as ANON and PUBLISHABLE are not in that list.
Free scan
CVE-2025-55182 package pin in the bundle
The free scan looks for an exact pin of react-server-dom-webpack, react-server-dom-parcel, or react-server-dom-turbopack at 19.0.0, 19.1.0, 19.1.1, or 19.2.0. A mention of the version in prose does not match. What the pin means.
Deep scan
Next.js middleware bypass probe (CVE-2025-29927)
A read-only header probe for the middleware bypass. It is not an exploit payload. How the check works.
Deep scan
Source maps that answer HTTP 200
The deep scan requests JavaScript source maps. A map that returns HTTP 200 for your own origin is reported, because it reconstructs original source. A third-party embed's map is informational.
Deep scan
Client-side auth and role checks
On a deep scan, JavaScript that gates a route or a role only in the browser is reported. Those checks are bypassable in devtools. A server check is the one that holds.
Deep scan
MCP tools/list with no credentials
If the origin exposes an MCP endpoint, the deep scan calls tools/list without credentials. A successful list is high, or critical when a tool name looks like SQL, shell, file, or admin. initialize alone is not a finding. The check does not call any tool. MCP checklist.
Deep scan, if you provide a config
MCP server definition review
Point a deep scan at an MCP server URL, a GitHub repo, or a config JSON and Sekrd reviews tool definitions for tool poisoning, tool shadowing, an over-permissioned filesystem, an exfiltration flow, hardcoded secrets, rug-pull drift against hashes you supply, and command injection when the source is a GitHub repo. This is separate from the unauthenticated tools/list probe.
Free scan
System prompt that contains a secret or an authorization rule
A systemPrompt, system_prompt, or systemMessage assignment in the bundle is reported when the prompt text contains credential-shaped material or tells the model to bypass auth or RLS. A prompt that merely mentions the word password is not enough.
Free scan
Payment webhook that accepts an unsigned POST
The payments check POSTs an unsigned body to a fixed list of Stripe, Paddle, and Lemon Squeezy paths. HTTP 200 with a non-HTML body, distinct from a catch-all, is critical. An HTML single-page-app shell is ignored.
What a scan does not do
A free scan reads the public URL: headers, HTML, and JavaScript. Supabase policy SQL, Firebase rules, TLS and DNS email checks, network ports, and the read-only active probes run on a deep scan. Deep scan is included with the $39 Pre-Launch Audit and with Continuous Pro. See pricing.
Active probes are reads and unsigned requests against known paths. Sekrd does not guarantee the app is safe, and a clean result is not a pentest. Findings the scanner could not observe are not listed.
Related reading
- How to secure a Cursor-built app
The deploy the agent cannot see from inside the editor.
- NEXT_PUBLIC_, VITE_, and EXPO_PUBLIC_
Which prefixes inline a secret into the client.
- CVE-2025-29927
The middleware bypass, checked without an exploit.
- MCP server security checklist
Tool poisoning, rug pulls, and token passthrough.
- MCP Inspector OAuth bug
October 21, 2026. Windows and WSL. Fixed in Inspector 2.12.0.
Frequently asked questions
Does Sekrd read my Cursor project?
Not unless you point a deep scan at a GitHub repository for the MCP review. The URL scan uses the deployed site only.
Is the MCP npm package the same thing as this scan?
No. The scan looks for an MCP server your app exposes. The @sekrd/mcp-server package, when published, lets an assistant start a Sekrd scan. It is documented on the docs page and is not on npm until the mcp-v* release tag is pushed.
Which Next.js issues are on the free scan?
The CVE-2025-55182 version pin is a regex on the bundle and runs on the free scan. The middleware bypass probe and source-map HTTP requests run on a deep scan.
How much does it cost?
Free within the quota on the pricing page. Pre-Launch Audit is $39 once and includes a deep scan. Continuous Pro is $24/month billed annually or $29/month billed monthly.