Loading ToolBench
MCP HTTP server exposing reverse engineering and APK analysis tools for AI agents. Provides APK parsing, SO file analysis, DEX extraction, security scanning, and binary patching capabilities.
This server has critical definition gaps. While all 20 tools are registered with names and basic descriptions, the descriptions are minimal (mostly 10-30 chars in English, with Chinese translations), parameter descriptions are sparse or missing, and input schemas lack proper type definitions and constraints. No output schemas are documented. Error handling is generic ('工具执行错误'). The tool set itself is concerning: 4 tools (patch_so_bytes, patch_so_string, nop_so_function, patch_dex_bytes, patch_dex_string) perform irreversible binary modifications with no confirmation, dry-run, or audit trail. Naming is inconsistent (some tools use underscores, descriptions mix Chinese/English). Per-tool scores average 38/100.
分析 APK 文件 (Parse and analyze APK file structure, manifest, permissions, and metadata)
分析 DEX 文件 (Parse DEX bytecode and extract classes, methods, and strings)
分析 SO 文件 (Parse ELF binary and extract symbols, sections, and metadata)
获取设备信息 (Get device information including manufacturer, model, Android version, SDK, ABI, and root status)
反汇编 SO 文件 (Disassemble native library function or address range)
转储文件十六进制 (Dump file contents in hexadecimal format)
Irreversible operations lack confirmation/dry-run. patch_so_bytes, patch_so_string, nop_so_function, patch_dex_bytes, patch_dex_string all modify binaries with no undo, confirmation step, or audit trail. An LLM mistake causes permanent data loss.
Descriptions are too short (10-40 chars) and lack context. Most tools have minimal English descriptions followed by Chinese translations. No guidance on WHEN to use each tool, WHAT it returns, or dependencies. LLMs cannot reliably select the right tool.
No output schemas documented. Tools return unstructured text (buildString, StringBuilder). LLMs cannot parse results reliably or chain tools. No field names, types, or structure guidance.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | <=2025-11-25 | v2 |
从 APK 提取并分析 SO 文件 (Extract native library from APK and perform ELF analysis)
提取 DEX 文件 (Extract DEX bytecode from APK)
获取应用 APK 路径 (Get the file path of an installed application's APK)
获取 Logcat 日志 (Retrieve device logcat output)
导入已安装应用 (Import installed application APK to workspace)
列出已安装应用 (List installed applications with package names, versions, and system app status)
列出 APK 中的 SO 文件 (List native library files in APK with ABI and size information)
列出工作区 APK (List APK files in the workspace directory)
NOP 函数 (Replace function with NOP instructions)
补丁 DEX 字节 (Patch DEX bytecode with raw bytes at specified offset)
补丁 DEX 字符串 (Replace string in DEX bytecode)
补丁 SO 字节 (Patch native library with raw bytes at specified offset)
补丁 SO 字符串 (Replace string in native library)
安全扫描 (Scan APK for security vulnerabilities and issues)
Parameter descriptions are missing or trivial. 'apk_path' has description 'APK 文件路径 (Path to APK file)', no format, no constraints, no examples. 'offset' and 'size' in dump_file_hex lack min/max bounds. LLMs guess at valid ranges.
Generic error handling. All exceptions caught and returned as 'Tool execution error: <message>'. No recovery guidance, no error classification (retryable vs fatal), no actionable next steps for LLM.