Skip to main content
MCP and security

MCP Description Scanner

Detect embedded prompt injection attacks and misleading capability descriptions in MCP tool definitions. Malicious MCP servers can inject adversarial instructions into tool descriptions that hijack LLM behavior. Scan before connecting any MCP server to a production agent.

Run MCP Description Scanner Interactively 100% client-side execution in your browser. Zero data retention.
Open Interactive Tool

What Problem MCP Description Scanner Solves

Production AI deployments require predictable token economics, rigorous parameter verification, and deterministic schema handling. MCP Description Scanner isolates parameters client-side before payloads hit third-party foundation model APIs.

Primary Production Use Cases

  • Audit third-party MCP server for description injection
  • Verify community MCP tools are safe to use

Security & Data Privacy Considerations

All computations execute locally in browser memory. No prompt payloads, test inputs, or secret credentials are ever retained or stored on central servers.

Want to automate this tool across multiple models? Orchestrate workflows in Agent Studio
Explore Agent Studio →