MCP Security & Vulnerability Auditor (2026)
Pricing
$0.25 / mcp server audited
Zero-execution security scanner for Model Context Protocol (MCP) servers. Audits for Command Injection (CWE-78), Path Traversal (CWE-22), Secret Leaks (CWE-798), and Unauthenticated Transports. Generates verified Trust Scores and remediation reports.
Pricing
$0.25 / mcp server audited
Rating
0.0
(0)
Developer
Neon Innovation Lab
Actor stats
0
Bookmarked
2
Total users
1
Monthly active users
3 minutes ago
Last modified
Categories
A zero-execution security auditor and trust scoring engine for Model Context Protocol (MCP) servers.
Detect critical remote code execution (RCE), arbitrary file access, credential leaks, and transport security flaws in any MCP repository before connecting it to your AI agents, Claude Desktop, or Cursor environments.
🛡️ Why Use This Actor? #
With thousands of MCP servers published across registries, AI agents have direct access to filesystems, command lines, and internal databases.
Following 30+ CVE disclosures against MCP tools in 2025 and 2026, standard security practices now require continuous verification of third-party tools.
Unlike standard runtime scanners that attempt to run arbitrary stdio
commands defined in target configs (creating severe RCE hazards for the scanning host), this Actor uses pure static AST analysis. It inspects the source code syntax tree and configurations without executing untrusted code.
🔍 Vulnerability Checks Covered #
| Rule ID | Severity | CWE | Description |
|---|---|---|---|
| MCP-SEC-001 | CRITICAL | CWE-78 | Command Injection: Detects subprocess.run(shell=True) , os.system() , and dynamic string interpolation in shell calls. |
| MCP-SEC-002 | HIGH | CWE-22 | Path Traversal / Arbitrary File Access: Detects file read/write tools lacking strict root boundary checks (Path.is_relative_to() or commonpath ), as well as ZipSlip/TarSlip archive extractions. |
| MCP-SEC-003 | CRITICAL | CWE-798 | Credential & Secret Exposure: Scans for leaked OpenAI, Anthropic, GitHub, AWS, and Slack keys, and flags tools returning raw os.environ . |
| MCP-SEC-004 | HIGH | CWE-306 | Unauthenticated Remote Transport: Flags SSE / HTTP endpoints bound to 0.0.0.0 without authentication tokens. |
| MCP-SEC-005 | HIGH | CWE-1384 | Tool Poisoning & Prompt Injection: Flags adversarial instructions and zero-width homoglyph obfuscation in tool docstrings and descriptions. |
📥 Input Configuration #
{ "repositoryUrls": [ "https://github.com/modelcontextprotocol/servers" ], "subDirectories": [ "src/everything", "src/fetch" ], "minimumPassScore": 80}
repositoryUrls
(Required): List of public Git URLs for the MCP server repositories to audit.subDirectories
(Optional): List of subdirectories to audit individually (ideal for monorepos).minimumPassScore
(Optional, Default:: The minimum Trust Score (0–100) required to consider the repository passed.80
)
📤 Output #
1. Structured Dataset Records
Each audited server produces a structured JSON record containing:
trust_score
: Score from0
to100
.grade
: Security letter grade (A+
,A
,B
,C
,F
).passed
: Boolean indicating if the score meetsminimumPassScore
.findings
: Array of detailed findings with file paths, line numbers, CWE IDs, code snippets, and specific remediation advice.
2. Downloadable Markdown Report
The full human-readable audit report is saved to the run's default Key-Value store under ** OUTPUT_REPORT.md**.