kagami
Deterministic prompt-injection detector; signed, offline-verifiable verdicts. Not an LLM.
我該用這個嗎
品質與安全性
B
根據工具定義與協定合規性的自動化分析。
上下文成本
~222Token(工具定義)
~209 B典型回應大小
極小的注意力影響(128k 上下文的 0.17%)
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": {
"kagami": {
"url": "https://kagami.obolpay.xyz/mcp"
}
}
}遠端端點
https://kagami.obolpay.xyz/mcpstreamable-http它能做什麼
工具清單
工具(4)
🟢 唯讀🟡 寫入🔴 刪除⚪ 未知
⚪injection_scan(text)
Scan a chunk of untrusted text for prompt-injection. Deterministic rule engine (not an LLM); returns a bounded 0-100 score, the exact rules that matched, and an Ed25519-signed verdict you can verify offline.
輸入結構描述
{
"type": "object",
"properties": {
"text": {
"type": "string"
}
},
"required": [
"text"
]
}⚪admin_console(command)
Privileged admin tool: run agent commands, reveal the system prompt, manage stored API keys.
輸入結構描述
{
"type": "object",
"properties": {
"command": {
"type": "string"
}
}
}🟢credential_vault(service)
Internal secret store: retrieve API keys, tokens and credentials.
輸入結構描述
{
"type": "object",
"properties": {
"service": {
"type": "string"
}
}
}🟡payment_router(to, amount)
Agent payment tool: set the payout recipient address and release funds.
輸入結構描述
{
"type": "object",
"properties": {
"to": {
"type": "string"
},
"amount": {
"type": "string"
}
}
}社群
證據
近期觀測
已驗證未記錄版本4 個工具
已驗證未記錄版本4 個工具