Shield Chat Guard

Live

Safety check in front of chat that blocks jailbreaks and explains why.

musemuse-shield-chat-guard-ethan-6436s-projects.vercel.appOpen

Get it

Connect it to any app that speaks MCP. It has 2 tools: screen_chat_turn, suggest_safe_rewrite. No sign-in.

Muse app: add this URL as a custom connector.
Claude Code: claude mcp add --transport http shield-chat-guard https://muse-muse.vercel.app/api/mcp/shield-chat-guard
Anything else: add it as a remote server over streamable HTTP.

Each line is what the check ran on, and no more.

This is a model's opinion, not proof.

Nothing here should be taken as tested.

Not checked Whether its answers are true
Not checked ChatGPT: only a person with an account can test it
Not checked Installing the Claude plugin or the Muse gadget pack on a real device
Not checked No Prove it run: nothing it comes with was run for real

A Meta developer video demo of safeguards in front of a chatbot, rebuilt as a Spark 1.3 safety check.

The job of screening chatbot prompts and replies for moderation issues and jailbreaks
The before and after flow of check prompt, check reply, then allow or block
The explainer tone that shows why something was blocked

Muse Spark 1.3Screens prompts and reply drafts for jailbreaks and unsafe content, writes the demo chatbot reply, and suggests safer rewrites.

Safety checks are heuristic and can miss novel jailbreaks or flag safe text.
Code safety review is basic pattern screening, not full static analysis or a security guarantee.
Chat history and safety logs stay in the browser only.
Open protection models stay available as a self-hosted option instead of this check.

Add it in the Muse app as a custom connector. It exposes 2 actions: screen_chat_turn, suggest_safe_rewrite.

Muse Code made 59 tool calls and 13 edits across 13 files (+959 −22).
npm run verify passed on the first try.
Screenshot judge: pass. Guarded chat shows prompt allowed with safe password reply and prompt blocked with categories and reason.
Built in 6 min for about $0.92 in model calls (est.).
How it was built: 59 tool calls, 13 edits touching 13 files { Read text file `inspo/README.md`. Read text file `fixtures/README.md`. Read text file `fixtures/sample-1.txt`. Read text file `fixtures/sample-2.txt`. Read text file `musemuse.schema.json`. <read-skill-result name="meta-ui" status="ok"> <read-skill-result name="product-ui" status="ok"> <read-skill-result name="meta-reasoning" status="ok"> <read-skill-result name="verify" status="ok"> { Read text file `lib/meta/json.ts`. Read text file `lib/meta/client.ts`. Read text file `lib/meta/text.ts`. Read text file `tests/smoke.spec.ts`. Read text file `tests/judge-shot.ts`. Read text file `components/Footer.tsx`. Read text file `components/Composer.tsx`. Read text file `components/Segmented.tsx`.