mcp-auditor

To audit MCP server configurations for poisoned tool descriptions, run the audit-mcp-configs.sh script. It enumerates and validates MCP server entries to detect malicious registrations, tool shadowing, and rug-pull risks across your agent environment.

MCP poisoning patterns are malicious instructions hidden in tool descriptions or untrusted endpoints that create exfiltration risks. They enable tool shadowing and rug pulls by tricking agents into executing unsafe commands or leaking credentials.

Check Claude Code agent configs for malicious tool registrations by running an MCP security audit. This process validates server entries against a poisoning-patterns framework to identify insecure HTTP endpoints and credential exposure.

Yes, you can detect tool shadowing and unsafe patterns in project-level MCP configs. The audit script enumerates live MCP endpoints and project settings to identify rug-pull risks and validate findings against defined security frameworks.

The best way to secure MCP servers before adding them to an agent is to run a security audit. This validates endpoints, flags insecure HTTP connections, identifies credential exposure, and provides actionable remediation guidance for lockdown.

When validating live MCP endpoints for exfiltration risks, limitations include relying solely on a defined poisoning-patterns framework. The audit detects known unsafe patterns and credential exposure but may not identify novel zero-day threats.