ServiceNow Action Fabric opens governed enterprise actions to AI agents

ServiceNow introduced Action Fabric at Knowledge 2026, describing it as a way for any AI agent to tap into secure, governed enterprise actions through ServiceNow’s generally available Model Context Protocol Server.

The core claim is that agents should not only read or write enterprise data. They should execute governed work: playbooks, approvals, catalog requests, workflows, and business rules, with controls routed through ServiceNow AI Control Tower.

Anthropic is named as an early design partner. ServiceNow describes a Claude Cowork scenario where Claude queries ServiceNow, identifies missing access items, asks the user to confirm, and routes the resulting requests through approval chains.

Why this matters

ServiceNow is making a platform bet: the valuable enterprise agent layer is not only model intelligence, but permissioned execution inside the systems that already run IT, HR, security, procurement, and customer operations.

If Action Fabric works as described, ServiceNow becomes the runtime for agents built elsewhere. Claude, Copilot, and homegrown agents could all route actions through the same governance layer.

Buyer take

Ask exactly which ServiceNow actions are exposed through MCP, which SKUs include access, how OAuth and role-based tool packages work, and how AI Control Tower logs agent activity.

The right pilot is not a demo chatbot. Test a workflow that requires identity, approvals, audit history, and failure handling, such as access provisioning or incident escalation.

What is still unclear

ServiceNow says its MCP Server is generally available and included in every Now Assist and AI Native SKU, with additional features due in the second half of 2026. Contract language and regional availability still need verification.