AIGovernrAIGovernrDocs
Ctrl K
Open workspace
Getting started
AIGovernr documentationQuickstartNavigating the appCore conceptsHow everything connects
Workspaces
CompaniesProjectsTeams & responsibilitiesDashboards & assignments
Account & administration
Plans & billingAccess control & rolesAccount securityNotifications
AI systems
AI SystemsRegister an AI SystemAI System classificationAI System Governance planGovernance plan vs PlaybookSystem documentation & testing recordsVendors & dependenciesShadow AI & asset graph
Obligations
Frameworks & obligationsRequirements & tasksGovernance knowledgeRegulatory updates
Assurance
Assure overviewControlsAssurance methodsTechnical evaluationsDPIAEvidence managementISO/IEC 42001 readinessGovernance maturity
Issues & actions
Findings vs risksCorrective actions
Decisions
ApprovalsApproval PoliciesDecision LedgerExceptions & reassessment
Continuous governance
MonitoringIncidentsAgent & MCP governance
Reusable governance
AI governance policiesPolicy templatesPlaybooks
Reporting
Reports, certificates & audit export
Client collaboration
Client portal & engagement
Guides
Run a client governance engagementOperate an AI governance programmeFramework implementation guides
Discovery tools
Website Compliance Scanner
AI assistance
AI Assistant
Developers
Integrations & APIs
Help
Troubleshooting & FAQ
DocsContinuous governance
Continuous governance

Agent & MCP governance

Define what an AI agent may do (tools, MCP servers, data, spend and human approval boundaries), check actions against that policy at runtime, and review redacted traces.

In the app
Project → Decide & Operate → Agent & MCP Governance

Set up an agent

  1. 1Create the agent in Asset Graph and assign an owner, then open Agent & MCP Governance.
  2. 2Complete the agent card: autonomy level, decision criticality, allowed and prohibited actions, allowed data and domains, spend and rate limits, the human approval boundary and the kill-switch owner.
  3. 3Add tool permissions for each tool, MCP server or delegate, with an effect of Allow or Deny.
  4. 4Use Policy check to simulate an action before going live.

Enforce at runtime

Create an enforcement key under Enforcement keys and copy it; it is shown once. Your agent or gateway calls the runtime authorisation API with the X-Agent-Enforcement-Key header before acting, and reports traces afterwards.

Saving a profile does not by itself prove enforcement. Enforcement happens only when your agent calls the policy service. Monitoring ingestion keys cannot be used as enforcement keys.

Last reviewed 2 October 2026

Next stepAI governance policiesAI governance policies are your organisation's official rules for governing AI, versioned and approved in each project.
IncidentsAI governance policies
On this page
Set up an agentEnforce at runtime
Capability status

Check current availability and limitations before relying on a feature.

View matrix