Check MCP resources, tool output and external text for suspicious instructions that may attempt to override policies, induce tool execution or extract protected information.
MCP tools and resources can introduce externally controlled text into an AI workflow. That text should be treated as untrusted data rather than automatically trusted instructions.
This scanner looks for common instruction-override, authority-impersonation, secret-exfiltration, tool-execution and safety-bypass patterns.
Scan MCP prompts, tool output, resources and external text for prompt injection patterns, instruction overrides, secret exfiltration and safety bypass attempts.
Use MCP Prompt Injection Checker as a focused pre-flight utility while building, reviewing or debugging an MCP integration. Paste the relevant configuration, protocol message, metadata, schema or value into the tool and review the returned findings before the same data reaches a production client or server.
Use the findings as a pre-deployment security review rather than as proof that an MCP server is safe. Security depends on runtime isolation, authorization, deployment policy and the behavior of the connected tools and systems.
A clean result means the input passed the rules implemented by this utility. It does not guarantee application security, protocol interoperability or correct business behavior. Test important integrations against the exact MCP client, SDK, gateway and server versions used in your deployment.