Add MCP (Model Context Protocol) security rules

[razashariff]

Summary

30 Semgrep rules for detecting security vulnerabilities in MCP (Model Context Protocol) server implementations. Covers JavaScript, TypeScript, and Python.

Rules

Category	Count	What it catches
Missing authentication	3	MCP endpoints without auth middleware
No input validation	3	Missing inputSchema, unvalidated tool arguments
Hardcoded secrets	3	API keys/tokens embedded in MCP configs
Binding to 0.0.0.0	2	Servers exposed on all interfaces
Missing TLS	4	Plaintext HTTP/WS, disabled cert verification
Command injection	4	exec/execSync, os.system, subprocess with shell=True
No rate limiting	2	Servers without rate limit middleware
Unsafe eval/exec	4	eval(), exec(), new Function() in handlers
Bonus (path traversal, SQLi, CORS, data leak, verbose errors)	5	Additional common MCP vulnerabilities

All rules include CWE codes, OWASP 2021 Top 10 references, and links to the OWASP MCP Security Cheat Sheet.

Why This Matters

• MCP has 97M monthly SDK downloads
• 30 CVEs filed in 60 days (early 2026)
• 492 servers exposed with zero auth (Trend Micro)
• No existing Semgrep rules for MCP security

References

• OWASP MCP Security Cheat Sheet Section 7
• MCP Specification

Source

Rules file: https://github.com/razashariff/agentpass-pentest/blob/main/semgrep-mcp-rules.yaml

Happy to submit a PR following your contribution guidelines.