mcp-detector

thilinawishvekeerthi/mcp-detector

★ 0 stars Jupyter Notebook 🤖 AI/LLM Updated today

A lightweight ML pipeline that detects prompt injection and tool poisoning attacks in MCP servers. It screens tool descriptions before they reach an LLM agent, with sub-30ms latency and per-decision explainability via TreeSHAP.

View on GitHub → Try with Claude — $10 free →

Quick Install

Copy the config for your editor. Some servers may need additional setup — check the README.

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "mcp-detector": {
      "command": "npx",
      "args": [
        "-y",
        "thilinawishvekeerthi/mcp-detector"
      ]
    }
  }
}

README Excerpt

**MCP Detector** is a lightweight, production-ready machine learning pipeline for detecting prompt injection and tool description poisoning attacks in Model Context Protocol (MCP) servers. It intercepts at the **tool registration boundary** — screening every MCP tool description before it reaches a language model agent — with sub-30 ms latency and full per-decision explainability via TreeSHAP.

mcp-detector

Quick Install

README Excerpt

Related AI/LLM Servers