Sign in
Catalog / Hands-on lab

Lab: Govern AI agents in the MCP catalog

Register an AI agent with an owner, add an MCP server to the catalog, allow only the tools it needs, decide approvals for a destructive tool and stop the agent with the kill switch. The MCP gateway service is a separate component that this lab does not install, so a small lab client, ticket-agent, reads the same catalog and approvals in its place.

Monosign University Pass 1 hAdvanced DE · EN · ES · TR Administrator Security analyst Developer

Version 6 · updated Nov 11, 2026What changed?

Hands-on lab
New
Sign in to start

Included with University Pass.

What you will learn

  • Give an agent an accountable owner, its own credential and the least permissions
  • Control which MCP tools an agent may use and which need a human decision
  • Stop an agent at once with the kill switch and confirm its credential no longer works

Your lab environment

YOUR LAB SERVERYour browserManagement and the webterminalManagementAI Security: Agents, MCPGateway, Approvalsticket-agentLab client in the web terminal,Support Assistant's keyTicketing MCP serversearch_tickets, add_comment,close_ticket, delete_customerMonosign APIAllowed tools, approvalrequestsRegisters, approves, killswitchDiscovers toolsCatalog andapprovals
Your lab environment: you govern the agent and the MCP catalog in Management, and the ticket-agent client stands in for the MCP gateway

Lab tasks

Each task is checked automatically against the real state of your lab environment.

  1. 1. Register and activate the Support Assistant agent with an owner
  2. 2. Give the agent a credential and three API permissions, then log the lab client in
  3. 3. Add the ticketing MCP server to the catalog, discover its tools and activate it
  4. 4. Allow search_tickets, add_comment and close_ticket, keep delete_customer off and assign the agent
  5. 5. Run close_ticket twice: approve one request and deny the other
  6. 6. Use the kill switch on the agent and confirm its next call is refused