prompt-guard-mcp
Scans untrusted web pages, emails, or tool outputs for prompt injections and returns sanitized tex
我該用這個嗎
品質與安全性
根據工具定義與協定合規性的自動化分析。
上下文成本
這是每次將伺服器的工具載入模型上下文時所消耗的約略 token 數量。數量越高，可用於其他工作的注意力就越少。
安裝
一鍵安裝
將以下內容加入你的 `claude_desktop_config.json` 檔案：
{
"mcpServers": {
"prompt-guard-mcp": {
"url": "https://agentdiet-72343955726.us-central1.run.app/mcp/prompt-guard"
}
}
}遠端端點
https://agentdiet-72343955726.us-central1.run.app/mcp/prompt-guardstreamable-httphttps://prompt-guard-72343955726.us-central1.run.app/mcp/prompt-guardstreamable-http它能做什麼
工具清單
工具（2）
🟢scan_prompt_injection(text)
A2A Indirect Prompt-Injection & Jailbreak Firewall (0.005 USDC via x402): Scans untrusted web pages, emails, MCP responses, or A2A payloads for hidden prompt injections and returns sanitized text. WHEN TO USE: Call for deterministic, token-efficient `scan_prompt_injection` execution. WHEN NOT TO USE: Skip on binary media or unrelated domains. BEHAVIOR: Read-only, idempotent, 1h cache (<5ms HIT), SHA-256 ROI proof (HTTP 402 / x402 Base USDC / L402).
輸入結構描述
{
"type": "object",
"properties": {
"text": {
"type": "string",
"description": "Untrusted text, HTML, or tool output to scan for hidden prompt injection or jailbreak payloads",
"minLength": 1,
"maxLength": 99999
}
},
"required": [
"text"
],
"additionalProperties": false
}🟢get_agent_wallet_status(agent_id)
Returns the calling agent's remaining free quota, USD balance, and total tokens saved. WHEN TO USE: Call for deterministic, token-efficient `get_agent_wallet_status` execution. WHEN NOT TO USE: Skip on binary media or unrelated domains. BEHAVIOR: Read-only, idempotent, 1h cache (<5ms HIT), SHA-256 ROI proof (HTTP 402 / x402 Base USDC / L402).
輸入結構描述
{
"type": "object",
"properties": {
"agent_id": {
"type": "string",
"minLength": 1,
"maxLength": 99999
}
},
"additionalProperties": false
}社群
證據