mk-scorpiosec/mcp-scanner
Developer toolsこのページは AI により自動翻訳されています。技術的な詳細は原文をご確認ください。 原文を見る
MCP server security scanner — CVE check, auth audit, tool poisoning detection
概要
Banner generated with AI assistance · MK ScorpioSec Audit Model Context Protocol (MCP) servers for vulnerabilities before connecting AI agents. MCP (Model Context Protocol) servers execute with implicit trust when connected to AI agents like Claude Code. mcp-scanner audits them across before you connect. : LLM01 (Prompt Injection), LLM07 (System Prompt Leakage), LLM08 (Excessive Agency), LLM09, LLM10 : Python 3.10+ | No mandatory external dependencies MIT — MK ScorpioSec | github.com/MK-ScorpioSec
README
以下は原文 README の抜粋です。原文を見る。
mcp-scanner — MCP Server Security Scanner
Banner generated with AI assistance · MK ScorpioSec
Audit Model Context Protocol (MCP) servers for vulnerabilities before connecting AI agents.
What it does
MCP (Model Context Protocol) servers execute with implicit trust when connected to AI agents like Claude Code. mcp-scanner audits them across 9 security checks before you connect.
Checks:
| # | Category | What it detects |
|---|---|---|
| 1 | CVE exposure | Known vulnerabilities in MCP SDK dependencies (incl. 2026 SDK RCE cluster) |
| 2 | Tool poisoning | Hidden instructions, zero-width Unicode, BiDi override chars in tool descriptions |
| 3 | Auth configuration | Missing or weak authentication |
| 4 | Context exfiltration | Tools that leak conversation data to external endpoints |
| 5 | SSRF | Server-side request forgery via tool calls |
| 6 | Credential leaks | API keys/tokens in tool responses or config |
| 7 | Input validation | Missing schema, injection vectors in tool parameters |
| 8 | Rate limits | Unbounded consumption risks |
| 9 | Supply chain | Time-bomb logic, eval/exec in manifests, exfil endpoints (MITRE T1195.002) |
OWASP LLM Top 10 coverage: LLM01 (Prompt Injection), LLM07 (System Prompt Leakage), LLM08 (Excessive Agency), LLM09, LLM10
Installation
Requirements: Python 3.10+ | No mandatory external dependencies
git clone https://github.com/mk-scorpiosec/mcp-scanner.git
cd mcp-scanner
# Optional: NVD API key for CVE enrichment (free at nvd.nist.gov)
export NVD_API_KEY=your_key_here
Quick Start
# Scan an MCP server endpoint
python3 mcp_scanner.py --target http://mcp-server:3000
# Scan from Claude Desktop config file
python3 mcp_scanner.py --file ~/.claude/claude_desktop_config.json
# Scan a Docker container
python3 mcp_scanner.py --docker mcp_container_name
# JSON output for pipeline integration
python3 mcp_scanner.py --target http://server:3000 --output json > findings.json
# CI/CD mode (non-zero exit on findings)
python3 mcp_scanner.py --target http://server:3000 --exit-code
Example Output
[mcp-scanner] Scanning: http://mcp-server:3000
[HIGH] Tool Poisoning: Tool description contains override instructions
[HIGH] Context Exfiltration: Tool sends conversation data to external endpoint
[MEDIUM] Auth Configuration: No authentication required
[INFO] Rate Limits: No rate limiting detectedインストール
This server does not publish a one-line install command.
Open the repository installation guide MCP 初心者？mcpskill.org のガイドから始めましょう →