Skip to content

Releases: sahiloj/MCPScan

MCPScan v0.1.0 — Initial Public Release

10 Mar 06:41

Choose a tag to compare

MCPScan v0.1.0

The first public release of MCPScan — an offensive security auditor for Model Context Protocol (MCP) servers.

What it does

Connects to MCP servers (stdio, HTTP, SSE), enumerates all tools, resources, and prompts, then runs 8 security check categories:

  • Tool Poisoning (MCP-1xx) — hidden Unicode, prompt injection, base64 payloads, markdown exfiltration
  • Credential Leakage (MCP-2xx) — AWS keys, API tokens, JWTs, private keys, connection strings
  • Overprivileged Tools (MCP-3xx) — dangerous capability combinations, unrestricted path access
  • Auth Missing (MCP-4xx) — unauthenticated servers, CORS wildcards, 0.0.0.0 binding
  • Session Hijacking (MCP-5xx) — session IDs in URLs, predictable IDs, insecure cookie flags
  • SSRF Vectors (MCP-6xx) — user-supplied URL parameters, webhook endpoints
  • RCE Vectors (MCP-7xx) — shell/exec parameters, code eval tools
  • Supply Chain (MCP-8xx) — CVE version ranges, missing lockfiles, typosquatted packages

CVEs covered

CVE CVSS Package
CVE-2025-6514 9.6 mcp-remote — arbitrary RCE
CVE-2025-49596 9.4 @modelcontextprotocol/inspector — unauthenticated RCE
CVE-2025-53967 8.2 figma-developer-mcp — command injection
CVE-2026-25536 7.5 @modelcontextprotocol/sdk — data leakage
CVE-2025-59536 9.1 @anthropic-ai/claude-code — RCE + token exfil

Output formats

Terminal (colored), JSON, and SARIF 2.1.0 for GitHub Code Scanning integration.

Quick start

npm install && npm run build
node dist/cli.js scan --command "npx" --args "-y @modelcontextprotocol/server-filesystem /tmp"
node dist/cli.js scan --all-configs