Loading ToolBench
MCP server for Panguard AI — control security from Claude/Cursor / 透過 Claude/Cursor 控制 Panguard 的 MCP 伺服器
All 8 tools share identical, minimal input schemas (single 'reason' string parameter) with no type constraints, enums, or validation guidance. Descriptions are present (100-150 chars each) and bilingual (English/Traditional Chinese), explaining WHAT and WHEN, but lack actionable parameter guidance. No output schemas are documented anywhere in the codebase. Tools follow verb_noun naming (checkIPHistory, checkUserPrivilege, etc.) but descriptions do not explain what data is returned or how to chain results. No error handling patterns, recovery guidance, or per-tool examples visible. Schema completeness is severely limited, all tools expose identical minimal input with no differentiation.
Check file hash and path against known malware signatures and suspicious patterns. Flags files in temporary directories (/tmp/, Temp) and files with suspicious extensions (.ps1, .vbs, .bat, .cmd, .scr, .pif, .hta). Use when the event involves file operations or process execution from a file path. / 檢查檔案雜湊和路徑是否符合已知惡意軟體特徵和可疑模式。標記在暫存目錄（/tmp/、Temp）中的檔案和有可疑副檔名的檔案（.ps1、.vbs、.bat、.cmd、.scr、.pif、.hta）。當事件涉及檔案操作或從檔案路徑執行程序時使用。
Check the geographic origin of a network connection by verifying the IP against baseline known connections. New IPs not in the baseline are flagged. Use when there is a source IP or remote address to verify. / 透過比對基線已知連線來檢查網路連線的地理來源。不在基線中的新 IP 會被標記。當有來源 IP 或遠端位址需要驗證時使用。
Check if an IP address has appeared in threat intelligence feeds or previous security events. Use when the event contains a source IP or remote address, especially for network-originated events. / 檢查 IP 位址是否出現在威脅情報來源或先前的安全事件中。當事件包含來源 IP 或遠端位址時使用，特別是網路來源事件。
Analyze network traffic patterns for anomalies by comparing against baseline connections. Flags new connections not in the baseline and connections to suspicious ports (4444, 5555, 8888, 1337, 31337, 6666, 6667, 9001). Use when the event involves network communication or outbound connections. / 透過與基線連線比較來分析網路流量模式的異常。標記不在基線中的新連線和連向可疑埠的連線（4444、5555、8888、1337、31337、6666、6667、9001）。當事件涉及網路通訊或外部連線時使用。
No output schemas documented. Tools return unstructured data; LLMs cannot plan downstream calls or extract required fields for chaining.
All 8 tools share identical input schema: single 'reason' string with no type constraints, enums, or validation. No guidance on what values are valid or expected.
Descriptions lack actionable parameter guidance. 'reason' parameter has no format constraints, length limits, or examples of what constitutes a valid reason.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2026-07-28+ | v2 |
Analyze the process ancestry chain for suspicious patterns. Checks if the process is known in the baseline and whether it has a suspicious parent process (e.g., cmd.exe, powershell.exe, bash, python). Use when the event is process-related or involves executable activity. / 分析程序祖先鏈以尋找可疑模式。檢查程序是否在基線中已知，以及是否有可疑的父程序（例如 cmd.exe、powershell.exe、bash、python）。當事件與程序相關或涉及可執行活動時使用。
Find correlated events within a 10-minute sliding time window. Correlates by source IP, username, process name, and event category. Use when you suspect the event is part of a larger attack pattern or campaign. / 在 10 分鐘滑動時間視窗內尋找關聯事件。按來源 IP、使用者名稱、程序名稱和事件類別進行關聯。當懷疑事件是更大攻擊模式或活動的一部分時使用。
Check if the event occurred during unusual hours compared to the environment baseline. Business hours are 06:00-22:00. Activities outside this window are flagged unless the user has a known night pattern. Use for any suspicious event to determine if timing is abnormal. / 檢查事件是否發生在相較於環境基線的異常時段。營業時間為 06:00-22:00。除非使用者有已知的夜間模式，否則此視窗外的活動會被標記。用於任何可疑事件以判斷時間是否異常。
Check if a user has elevated privileges (root/admin) or is an unknown user not in the baseline. Use when the event involves authentication, authorization, or any user-attributed activity. / 檢查使用者是否擁有提升的權限（root/admin）或是基線中不存在的未知使用者。當事件涉及認證、授權或任何使用者相關活動時使用。
No error handling patterns or recovery guidance. Tools do not indicate what errors are retryable, user-fixable, or fatal.
Tool composition unclear. Descriptions explain WHEN to use each tool but not how results chain together or what IDs/references are returned for downstream calls.