jshookmcp

mcp
Guvenlik Denetimi
Basarisiz
Health Gecti
  • License — License: AGPL-3.0
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Community trust — 1291 GitHub stars
Code Basarisiz
  • rm -rf — Recursive force deletion command in .github/workflows/publish-packages.yml
  • fs module — File system access in .github/workflows/publish-packages.yml
Permissions Gecti
  • Permissions — No dangerous permissions requested
Purpose
This is a Model Context Protocol (MCP) server that equips AI models with a comprehensive suite of tools for JavaScript analysis, browser automation, and security reverse engineering. It enables deep inspection of web applications directly from compatible AI clients like Claude Desktop or Cursor.

Security Assessment
This tool requires extensive access to sensitive system resources to function. By design, it performs heavy network monitoring, browser automation via the Chrome DevTools Protocol, and local process inspection. It does not request highly dangerous OS-level permissions directly, but its nature implies significant local execution capabilities.

The automated code scan flagged a `rm -rf` (recursive force delete) command and general file system access within its CI/CD workflows (`.github/workflows/publish-packages.yml`). While common in build and publishing scripts, these commands still pose a theoretical risk if the workflow is compromised. No hardcoded secrets were detected. Overall risk is rated as Medium due to its inherent need to interact with local browsers, debug running processes, and intercept network traffic.

Quality Assessment
The project demonstrates strong health and high community trust. It is licensed under AGPL-3.0 and is highly active, with its most recent code push occurring today. With over 1,200 GitHub stars, it has clearly established a solid user base. Furthermore, it provides comprehensive documentation and quick-start guides for easy setup.

Verdict
Use with caution—while the project is high-quality, popular, and actively maintained, granting any application this level of deep system, network, and browser access carries inherent risks that users should carefully consider.
SUMMARY

js hook toolkit that all you need

README.md

@jshookmcp/jshook

License: AGPLv3
Node.js 20.19+ or 22.12+
TypeScript
MCP
pnpm

English | 中文

An MCP (Model Context Protocol) server with a runtime-registry-driven catalog of built-in tools for AI-assisted JavaScript analysis and security analysis. It combines browser automation, Chrome DevTools Protocol debugging, network monitoring, intelligent JavaScript hooks, LLM-powered code analysis, process and memory inspection, WASM tooling, source-map reconstruction, AST transforms, and composite workflows in a single server.

Documentation / Quick Links

🚀 Quick Start

Use jshookmcp instantly with Claude Desktop or Cursor without installing anything globally.

Claude Desktop Configuration (claude_desktop_config.json):

{
  "mcpServers": {
    "jshook": {
      "command": "npx",
      "args": ["-y", "@jshookmcp/jshook@latest"],
      "env": {
        "JSHOOK_BASE_PROFILE": "search"
      }
    }
  }
}

(Note for Windows users: If npx is not found, specify the absolute path to npx.cmd)

🌟 Key Highlights

  • 🤖 AI-Driven Analysis: Leverage LLMs for intelligent JavaScript deobfuscation, cryptographic algorithm detection, and AST-level code comprehension.
  • Search-First Context Efficiency: BM25-powered search_tools + dynamic boosts cut jshook's tool-schema init delta from ~40.0K+ tokens (full) to ~3.0K (search) (Claude server-side count; excludes Claude Code base prompt).
  • 🎯 Progressive Capability Tiers: Three built-in profiles (search/workflow/full), with search as the default base tier for on-demand capability scaling.
  • 🌐 Full-Stack Automation: Seamlessly orchestrate Chromium/Camoufox browsers, CDP debugging, and network interception as atomic actions.
  • 🛡️ Advanced Anti-Debug: Built-in evasion for debugger statements, timing checks, and strict headless bot fingerprinting techniques.
  • 🧩 Dynamic Extensibility: Hot-reload plugins and workflows from local directories without recompiling the core server.
  • 🔧 Zero-Wiring Extensibility: Auto-discovered domains via manifest.ts, lazy handler instantiation, and B-Skeleton contracts for plugins/workflows.
  • 🛠️ Reverse Engineering Toolchain: Integrated WASM disassembly, binary entropy analysis, in-memory scanning, and bridges for Burp Suite/Ghidra/IDA Pro.

🛡️ Core Capabilities

JSHookMCP exposes 360+ atomic tools across 36 domains, empowering AI orchestrators with unparalleled capabilities:

  • 🕸️ Browser Automation & Reverse Engineering: Zero-config Chromium/Camoufox injection, CDP (Chrome DevTools Protocol) orchestration, and iframe evaluation bypasses.
  • 📡 Network Interception & Spoofing: Deep HTTP/2 frame building, MiTM traffic capture, GraphQL introspection, and Burp Suite bridge.
  • 🧠 AST & Semantic Analysis: LLM-powered deobfuscation, WebAssembly (WASM) disassembly, Source Map reconstruction, and binary entropy visualization.
  • 🧰 Process & Memory Forensics: Native Frida instrumentation, memory scanning, pointer dereferencing, and strict Anti-Debug mitigation.
  • 🔌 Dynamic Extensibility: Hot-reloadable B-Skeleton plugins and declarative WorkflowContract pipelines.

View the complete 36-domain tool catalog ↗

Architecture & Performance

[!TIP]
Context Efficiency Benchmark: Built-in tool-schema init delta (Claude server-side count): search ≈ 3.0K tokens vs full ≈ 40.0K+ tokens.

  • Progressive Tool Discovery: search_tools meta-tool (BM25 ranking) + activate_tools / activate_domain + profile-based tier upgrades (boost_profile)
  • Search-tier behavior: search_tools only searches and ranks results; it does not auto-run activate_tools, and it does not auto-run boost_profile. Preferred chain: search_tools -> activate_tools / activate_domain -> boost_profile only when needed
  • Do not boost for one tool: activate_tools can register exact tools across tiers from the current base tier; boost_profile is better when you expect to reuse a broad family of related tools repeatedly
  • Lazy Domain Initialization: Handler classes instantiated via Proxy on first invocation, not during startup
  • Domain Self-Discovery: Runtime manifest scanning (domains/*/manifest.ts) replaces hardcoded imports; add new domains by creating a single manifest file
  • B-Skeleton Contracts: Extensibility contracts for plugins (PluginContract), workflows (WorkflowContract), and observability (InstrumentationContract)
  • MCP ToolAnnotations: Every tool carries semantic annotations (readOnlyHint, destructiveHint, idempotentHint, openWorldHint) enabling AI orchestrators to reason about tool safety and side-effects before invocation

Registry Snapshot

The built-in surface below is generated from the runtime registry and checked in CI.

  • Package version: 0.2.9
  • Built-in Tools: 364
  • Domains: adb-bridge, antidebug, binary-instrument, boringssl-inspector, browser, canvas, coordination, core, cross-domain, debugger, encoding, evidence, extension-registry, graphql, hooks, instrumentation, macro, maintenance, memory, mojo-ipc, network, platform, process, protocol-analysis, proxy, sandbox, shared-state-board, skia-capture, sourcemap, streaming, syscall-hook, trace, transform, v8-inspector, wasm, workflow
  • Note: this snapshot is generated from the runtime registry; do not edit the counts by hand.

View the complete Tool Reference ↗

Project Stats

Star History

Star History Chart

Activity

Yorumlar (0)

Sonuc bulunamadi