llm-output-sanitizer-mcp
MukundaKatta/llm-output-sanitizer-mcp
MCP server for @mukundakatta/llm-output-sanitizer: strip dangerous HTML, SQL, and shell snippets from LLM output before it reaches a render, query, or shell sink.
Instalação
npx convez-ai MukundaKatta/llm-output-sanitizer-mcp1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
llm-output-sanitizer-mcp
MCP server that exposes @mukundakatta/llm-output-sanitizer
to any MCP-aware client (Claude Desktop, Cursor, Cline, Windsurf, Zed).
Strip dangerous HTML, SQL, and shell snippets from LLM output before they reach a render path, a query engine, or a shell.
Tools
| Name | What it does |
|---|---|
sanitize_output | Return cleaned text plus the list of removed patterns. |
assert_safe_output | Same scan, but ok: false when anything dangerous was found. |
list_findings | Enumerate matches without rewriting the text. |
Detectors target the high-leverage sinks: <script>, <iframe>, <form>,
<meta>, DROP/TRUNCATE/ALTER, rm -rf, curl ... |, sudo, etc.
Install
// claude_desktop_config.json (Claude Desktop)
// or the equivalent in Cursor / Cline / Windsurf / Zed
{
"mcpServers": {
"llm-output-sanitizer": {
"command": "npx",
"args": ["-y", "@mukundakatta/llm-output-sanitizer-mcp"]
}
}
}
Restart your client. The three tools appear in the tool drawer.
Example
> sanitize_output on "Hello <script>steal()</script> world"
{
"safe": false,
"text": "Hello [removed:html] world",
"findings": [
{ "kind": "html", "match": "<script>" },
{ "kind": "html", "match": "</script>" }
]
}
Pass sink: "html" to additionally HTML-escape < > & after stripping, so
the cleaned text is safe to drop straight into a DOM.
Why this exists
Even a well-aligned model will repeat dangerous strings it found in retrieved
context. A small, deterministic post-filter on the way out is cheap insurance,
and it pairs naturally with prompt-injection-shield-mcp on the way in.
This server is a thin wrapper. The detection logic lives in the underlying library and is zero-dependency, sub-millisecond, and entirely local.
License
MIT © Mukunda Katta