app.vercel.cloakcheck-wheat/cloakcheck
app.vercel.cloakcheck-wheat/cloakcheck · 1.0.0
Scan a page for hidden prompt-injection payloads targeting AI agents.
Nobody here has read this server's code, because it publishes none. The description above is the maker's own, from the registry. A connection test shows whether it answers and what tools it says it has; it never calls a tool, so it cannot show what one does with your data.
Where it sits in the directory's order
What can be seen from outside, weighed as the directory publishes.
Observable signals: 45 of 100 points (at most 60 here)
| Input | Points | What was seen |
|---|---|---|
| Source published | held at 0 | No entry here publishes source: that is what puts it in this directory. |
| Licence stated | held at 0 | With no files, there is nothing for a licence to be stated in or held against. |
| Advisory state | held at 0 | Advisory databases index packages, and these entries publish none, so there is nothing to look up. |
| Auth declared | 0 of 15 | The registry entry declares no key. That is what it declares, not a finding that it has no protection. |
| Latest connection test reached it | 20 of 20 |
Where it answers
https://cloakcheck-wheat.vercel.app/api/mcpstreamable-http
Connection test
The MCP handshake, then a request for the tool list, with no key and no data of yours. Run nightly, and by anyone, at most once every ten minutes per server.
It completed the handshake and listed 1 tool.
- Protocol
- 2025-03-26
- Calls itself
- cloakcheck 1.0.0
- Handshake time
- 87 ms
- HTTP status
- 200
The tools it lists (1, as of 6 days ago)
The tool-description rules found nothing in these descriptions. They look for instructions aimed at a model and for hidden characters; they cannot see what a tool does when it runs.
check_page_for_injectionScan a single web page for content planted to hijack an AI agent reading it -- invisible unicode (zero-width chars, the unicode 'tag' block used for steganographic prompt injection), CSS-hidden instruction text, and instruction-shaped language in alt/title/aria-label attributes a human would never read. Does NOT judge whether visible body text is safe -- only content hidden from normal human reading flow is flagged, so a page that legitimately discusses prompt injection won't false-positive on itself. Call this before an autonomous shopping/browsing agent acts on a page's content (add to cart,
How this entry becomes a listing
For the maker. The catalogue lists what it can read, and this server publishes nothing to read yet. There are two routes, and only the first is open today.
Publish the source Open today
- Put the server's source in a public repository on github.com, with a licence.
- Keep a server.json in that repository naming this server,
app.vercel.cloakcheck-wheat/cloakcheck, and declare the repository in it:"repository": { "url": "https://github.com/owner/repo", "source": "github" }, adding"subfolder"when the server lives in a folder. - Publish that version to the official MCP registry.
- The nightly registry sweep records the declaration. This page stays, dated, and says the source is declared but not yet read.
- The catalogue reads declared repositories at a pinned commit, in batches run by hand, and lists the servers that meet the batch's rules (among them a server.json at that commit naming the server, an https endpoint and a licence). When it lists this server, this page links to the listing. There is no schedule, so no date can be promised.
List it through the maker studio Not open yet
From the official MCP registry, last updated there 25 days ago. The registry entry · cloakcheck-wheat.vercel.app