AgentDiet A2A Indirect Prompt-Injection & Jailbreak Firewall for ChatGPT

dev.agentdiet/prompt-guard

Scans untrusted web pages, emails, MCP responses, or A2A payloads for hidden prompt injections an.

client:ChatGPT transport:streamable-http tools:2

Install AgentDiet A2A Indirect Prompt-Injection & Jailbreak Firewall in ChatGPT

Settings → Connectors → Advanced → Developer mode

https://prompt-guard.agentdiet.dev/mcp

Custom connectors are a paid-plan feature and take a remote URL only. ChatGPT cannot run a packaged server locally. ChatGPT docs