FAQs

Frequently Asked Questions About Invigilate

Answers to common setup, compatibility, and security questions for teams implementing MCP proxy controls.

FAQ

MCP proxy and security answers

What is Invigilate?

Invigilate is an MCP proxy and security middleware for AI agents. It sits between your agent and MCP server to apply tool filtering, policy rules, schema sanitization, notifications, and audit logging.

Do I need code changes to adopt it?

No. You register your MCP server once and then point your agent to the generated proxy URL. The proxy stays JSON-RPC compatible with standard MCP flows.

How does Invigilate improve MCP security?

You can classify tools by risk, define condition-based rules, and block destructive calls before they reach upstream systems. Notifications and logs provide an additional review layer.

Can I block destructive AI agent calls?

Yes. Set rule actions to Block or Block + Notify and match by tool_risk or argument patterns like SQL or shell command indicators.

Which providers are supported?

Invigilate supports provider-specific schema output modes for major LLM ecosystems, including Bedrock, OpenAI-style function schemas, Anthropic, and Vertex-oriented formats.

What is captured in MCP audit logs?

Logs capture tool names, outcomes (allowed, blocked, error), durations, optional model/agent context, matched rule name, timestamps, and error details where relevant.