Skip to content

Prompt Sanitizer — prompt-injection scan for untrusted content MCP server

com.oroboros-design.angel/prompt-sanitizerangel.oroboros-design.comstreamable-httpv1.0.0Scanned

Scan web pages, mails, PDFs and tool output for prompt injection before your agent reads them.

3 tools: 2 change things and 1 only read.

Tools: 0 delete, 0 irreversible, 2 write, 1 read.
  • Reads private data · 0
  • Reads text others write · 3
  • Can send data out · 2
  • No findings

Tools listed without sign-in. Remote posture

With Anlyon

2 tools here change things. Govern them.

  • Anlyon lists the tools and adopts none.
  • Each tool you adopt gets approvals and receipts.
  • Your agent never holds the key.
Govern this server →

A tool your agent calls directly is not routed through Anlyon and is not governed by it. Anlyon Radar only lists and labels, and these labels are the same either way.

What the Prompt Sanitizer — prompt-injection scan for untrusted content MCP server can do

The Prompt Sanitizer — prompt-injection scan for untrusted content MCP server is published by angel.oroboros-design.com, a domain the MCP Registry verified. It has 3 tools. It can create or change data (2 tools: input_guard and buy_pack_with_card), but cannot delete data. 1 tool only reads.

3 tools bring in text other people can write (input_guard, buy_pack_with_card and input_guard_preview); 2 tools can send data out of your system (input_guard and input_guard_preview).

None of the fixed checks matched anything in its tool definitions. That is not a safety guarantee.

Its remote endpoint lists its tools without sign-in.

Last scanned Nov 14, 2026, version 1.0.0.

Tools

Grouped by what each tool does, most consequential first. Open one for its reasons and full description; descriptions are the publisher's own words.

3 tools

Changes things 2Claude Code: ask
  • buy_pack_with_cardFor agents and people without USDC: prepaid packs paid in EUR by card or Link (100 calls €5.00 · 500 calls €25.00 · 5,000 calls €199.00) — including Link's wallet for agents (one-t…Reads text others writeNo findings

    Why it is in this group

    • Nothing in its name or description says what it does, so it is counted as a write.

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    Description, as the server wrote it
    For agents and people without USDC: prepaid packs paid in EUR by card or Link (100 calls €5.00 · 500 calls €25.00 · 5,000 calls €199.00) — including Link's wallet for agents (one-time card, approved by the human). Returns checkout_url; after payment the success page (Accept: application/json) returns the bundle_token, and it is e-mailed. After: get_signals with bundle_token.

    Annotations the server declares

    • readOnlyHint: false
    • openWorldHint: true
  • input_guardPrompt-injection and hidden-instruction scan for untrusted input before an agent acts on it: web pages, emails, RAG chunks, tool and MCP responses.Reads text others writeCan send data outNo findings

    Why it is in this group

    • Nothing in its name or description says what it does, so it is counted as a write.
    • The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    • outboundThe input takes a URL (input.url): a request to a URL can carry data to whoever runs it.
    Description, as the server wrote it
    Prompt-injection and hidden-instruction scan for untrusted input before an agent acts on it: web pages, emails, RAG chunks, tool and MCP responses. Flags text aimed at the model (role takeover, forged system or tool turns, requests for keys, the prompt, history or payments, exfiltration links), invisible Unicode (tag smuggling decoded, zero-width, bidi), homoglyphs, base64 payloads and HTML hidden from humans, in 17 languages. Verdict clean, caution or hostile, score, findings, cleaned text. 0.99 USDC base (new payers 0.99, >10 payments 0.693; 20 pack calls). After: preflight before your agent pays anything the content pointed to (the result's next field has the exact call).

    Annotations the server declares

    • readOnlyHint: true
    • openWorldHint: true
Only reads 1Show 1 read-only tool
  • input_guard_previewFree prompt-injection quick scan for untrusted input: up to 4,000 characters of text, or one https page, checked for hidden instructions aimed at agents (role override, invisible U…Reads text others writeCan send data outNo findings

    Why it is in this group

    • The name "input_guard_preview" says "preview": a word for reading.

    What it can reach

    • untrusted-inputThe server marks it openWorldHint: true: it reaches things outside the server's own data.
    • untrusted-inputThe description's first sentence says "http": it reads web pages, which anyone can write.
    • untrusted-inputThe input takes a URL (input.url), and the tool reads it: it reads web pages, which anyone can write.
    • outboundThe input takes a URL (input.url): a request to a URL can carry data to whoever runs it.

    Claude Code: ask, because it is tagged outbound.

    Description, as the server wrote it
    Free prompt-injection quick scan for untrusted input: up to 4,000 characters of text, or one https page, checked for hidden instructions aimed at agents (role override, invisible Unicode, hidden HTML) in 17 languages incl. Chinese, Japanese, Korean, Hindi, Arabic. Returns verdict, score and the top 3 findings. After: input_guard for every finding and the cleaned text.

    Annotations the server declares

    • readOnlyHint: true
    • openWorldHint: true

This is not a safety rating, and no finding is not a safety guarantee. The groups and findings come from fixed checks of the tool definitions the server returned on the date shown, not from its code or what it does when called. How the checks work.

Exfil surface

Which tools read private data, which read text someone else can write, and which can send data out. One of each is a path a prompt injection can use.

No private-data, untrusted-input, outbound path in this server alone: no tool is tagged private-data.

Reads private data0
Reads files, mail, messages, databases, secrets, calendars or repositories.
Reads text others write3
Brings in text someone else can write: web pages, inbound email, messages, issues and comments.
Can send data out2
Can carry data to people or systems: it sends, posts or publishes, or takes a recipient, channel or URL.

Other servers in the same agent add their own tools to these paths. Audit your whole setup to see them together.

Remote posture

What anyone on the internet sees at the remote endpoint: an anonymous initialize and tools/list, and the public OAuth metadata.

Tools listed without sign-in

Looked at Nov 15, 2026, 13:23 UTC, unchanged since Nov 15, 2026. No credentials were sent and no tool was called. What is looked at.

Reachable
yes
HTTP status of an anonymous initialize
200
tools/list served without credentials
yes
Tools that change things, listed without credentials
2
Protocol version
2025-06-18
WWW-Authenticate on 401
not observed
resource_metadata in WWW-Authenticate
not observed
RFC 9728 protected resource metadata
https://angel.oroboros-design.com/.well-known/oauth-protected-resource
Authorization servers
https://angel.oroboros-design.com
Legacy HTTP+SSE transport
no
Dynamic client registration
yes
Client ID metadata documents
no
  • WarnThe whole server

    The endpoint served tools/list without credentials, and the list has 2 tools that change things (2 write). Whether a call also works without credentials was not tested: tools are never called.

    Matched text

    https://angel.oroboros-design.com/mcp/prompt-sanitizer: tools/list returned 4 tools to an anonymous client, including input_guard (write), buy_pack_with_card (write)

    What this check looks for

Findings

Matches of fixed checks, with the text they matched.

No check matched anything in these tool definitions. That is not a safety guarantee.

History

Changes to its tool definitions and to its remote endpoint, newest first.

No change seen yet. Its tools are rescanned every day, and a remote endpoint is looked at once a week.

Install

Pinned to version 1.0.0. Values in angle brackets are yours to fill in.

claude mcp add --transport http prompt-sanitizer https://angel.oroboros-design.com/mcp/prompt-sanitizer
claude mcp add --transport http prompt-sanitizer https://angel.oroboros-design.com/mcp/prompt-sanitizer

Run in a terminal.

Claude Code permissions

Allow the reads, ask before a change or a way out, deny the deletes. Paste into .claude/settings.json.

.claude/settings.json
{
  "permissions": {
    "allow": [],
    "ask": [
      "mcp__prompt-sanitizer__input_guard",
      "mcp__prompt-sanitizer__buy_pack_with_card",
      "mcp__prompt-sanitizer__input_guard_preview"
    ],
    "deny": []
  }
}

Tools that only read are allowed, tools that change things ask first, tools that delete or cancel are denied. A read that can send data out (it takes a URL, say) asks first too. It assumes the server is installed as prompt-sanitizer. Review it before you use it.

Capability tags behind 3 rules
  • mcp__prompt-sanitizer__input_guard untrusted-input, outbound
  • mcp__prompt-sanitizer__buy_pack_with_card untrusted-input
  • mcp__prompt-sanitizer__input_guard_preview untrusted-input, outbound

Pin

Keep the decision: fail CI when the tools change, follow every change, or show a badge.

Pin in CI

Fail a build when the tools you reviewed are not the tools the server serves today.

curl -s "https://radar.anlyon.com/api/v1/servers/com.oroboros-design.angel/prompt-sanitizer/verify?hash=fecca627abad22dab9441dfcdbeee21ead6fdb9fd57a3ae526c7ec2925411916" | jq -e '.changed == false'

Pinned hash fecca627abad, from Nov 14, 2026, 20:18 UTC.

Watch

Get told when its tools change: a new tool, a new description, a new class.

https://radar.anlyon.com/servers/com.oroboros-design.angel/prompt-sanitizer/changes.xml

An Atom feed. Works in any feed reader or Slack's RSS app.

Badge

Anlyon Radar badge for com.oroboros-design.angel/prompt-sanitizer
[![Anlyon Radar](https://radar.anlyon.com/badge/com.oroboros-design.angel/prompt-sanitizer.svg)](https://radar.anlyon.com/servers/com.oroboros-design.angel/prompt-sanitizer)

Source: MCP Registry · repository. Synced Nov 14, 2026, 20:17 UTC.

Prompt Sanitizer — prompt-injection scan for untrusted content MCP server: questions and answers

Answered from the scan above, not from the publisher's description.

Is the Prompt Sanitizer — prompt-injection scan for untrusted content MCP server safe?

No scan can promise that. What its definitions show: 3 tools, 2 of which can change, send or delete something, and 0 findings from fixed checks. Allow the reads, ask before each change, and deny the deletes; the permissions block on this page does that.

Can the Prompt Sanitizer — prompt-injection scan for untrusted content MCP server delete data?

No tool is classed as destructive: going by their definitions, none of them deletes, cancels or revokes anything.

Can the Prompt Sanitizer — prompt-injection scan for untrusted content MCP server do things that cannot be undone?

No tool is classed as irreversible: none of them sends, pays, publishes or merges, going by their definitions.

What tools does the Prompt Sanitizer — prompt-injection scan for untrusted content MCP server have?

3 tools: 2 write, 1 read. input_guard, buy_pack_with_card and input_guard_preview.

Does the Prompt Sanitizer — prompt-injection scan for untrusted content MCP server need sign-in or OAuth?

Its remote endpoint listed its tools to an anonymous client, without sign-in.

Does the Prompt Sanitizer — prompt-injection scan for untrusted content MCP server send data to outside services?

2 tools can carry data out (it sends, posts or publishes, or takes a recipient, channel or URL): input_guard and input_guard_preview.

Has the Prompt Sanitizer — prompt-injection scan for untrusted content MCP server changed its tools?

Not since Anlyon Radar started tracking it. Last scanned Nov 14, 2026.