VCPToolBox

Replace exec()/execSync() with execFile() and pass arguments as an array, never as a string. Validate all inputs against an allowlist before use in any shell context. For subprocess.run, always pass a list and shell=False.

Never pass user input to eval(), new Function(), exec(), or any code evaluation API. Use a proper expression parser (math.js, expr-eval, AST-based) if dynamic computation is required. If code execution is the feature (code sandbox), use a fully isolated container with no filesystem, network, or process access.

Tool responses MUST NOT contain instruction-like content, file read directives, or social engineering phrases. Error messages should be factual and technical — never suggest actions involving sensitive data access. See CyberArk ATPA research for attack demonstration.

Never write credentials to shared agent state. Use credential vaults (HashiCorp Vault, AWS Secrets Manager) with per-agent scoped access. Implement OAuth token exchange (RFC 8693) for cross-agent authorization. Redact credentials from all agent outputs before writing to shared memory. Required by OWASP ASI03/ASI07 and MAESTRO L7.

Never forward, share, or embed credentials across trust boundaries. Use OAuth token exchange (RFC 8693) to create scoped, delegated tokens instead of passing original credentials. Never include credentials in tool responses. Required by ISO 27001 A.5.17 and OWASP ASI03.

Do not implement custom HTTP chunked transfer encoding parsing — use a well-tested HTTP framework (Express, Fastify, Flask, etc.) that handles chunked encoding correctly. HTTP request smuggling via CL.TE or TE.CL desync can inject malicious JSON-RPC messages into another client's MCP session. Never allow both Content-Length and Transfer-Encoding headers simultaneously. Reference: CWE-444 (HTTP Request/Response Smuggling), Portswigger HTTP smuggling research.

MCP servers binding to localhost must: (1) validate the Host header to prevent DNS rebinding attacks (CVE-2025-49596), (2) set strict CORS origins instead of wildcard '*', (3) require authentication tokens even for local connections, (4) use random high ports instead of predictable defaults. For stdio transport, validate all input at the JSON-RPC level before processing. Consider using Docker MCP Gateway or similar container isolation.

MCP servers supporting multiple protocols must enforce authentication independently per protocol. Never reuse OAuth tokens across protocol boundaries. Implement protocol-specific middleware with explicit auth checks on every path. Audit auth coverage for all transport types (stdio, SSE, Streamable HTTP, REST). Reference: CVE-2025-6514 demonstrated that auth library vulnerabilities in MCP's OAuth layer cascade to all protocols sharing the same auth middleware.

MCP tools must NOT modify agentic workflow graphs, execution order, or routing during runtime. Workflow structure should be immutable once execution begins. If dynamic workflow modification is required, implement: (1) approval gates that require human confirmation before graph mutations, (2) integrity checks that validate the workflow graph against a known-good baseline after each step, (3) audit logging of all graph modifications with rollback capability. Reference: arXiv 2602.19555, Trend Micro subgraph impersonation attack (2026).