ServiceNow's MCP plugin is listed in OpenAI's marketplace. The October 28 announcement describes planned incident lookup, change-risk assessment and record updates, without specifying current availability.

Customers would connect an instance and choose capabilities. The companies say Action Fabric will apply existing rules and permissions, with ServiceNow executing actions.

The big change

  • The intended connection would expose ServiceNow workflows in ChatGPT under customer rules and permissions. A response in chat and a confirmed record change are separate things to check.
  • An offered tool is not the same as permission to change a particular record. Administrators need to understand both layers before enabling an operation.
  • The marketplace plugin and the richer on-call preview are different release stages. Future timelines are plans, and an announcement alone is not evidence of results in a particular organization.

What happens when an action is requested

The Model Context Protocol, or MCP, gives an AI application a way to connect to external tools. A conversational request and the action it asks for are separate parts of the workflow: a user needs access to the relevant operation before the system can execute it.

The announcement does not identify every available tool. BIG CHANGE reviewed the companies' statements and supporting documentation; it did not connect a tenant or perform an update.

An AWS and ServiceNow technical guide illustrates the distinction in a separate integration. It describes an MCP Server Console where administrators configure exposed tools, with tool scoping, access control lists and role masking. Its instructions address AWS DevOps Agent and its own prerequisites. It shows why the list of exposed operations and the permissions behind them are separate things to inspect.

For an administrator preparing a rollout, an account that can read an incident and an account that can update one do not necessarily have the same authority. The useful check is whether the capabilities offered to a role match the work that role is supposed to perform. An answer in a conversation and a successful record update are also different outcomes; the latter should be confirmed in the system of record.

What is available and what is planned

Enhanced capabilities are expected by end-2026. An illustrative preview has an on-call engineer reviewing alerts, changes and affected services.

Security capabilities are expected in early access in January 2027 and general availability in the second quarter. Voice risk assessments are expected in the first quarter, and Codex in AI Workflow Factory during 2027. The announcements do not provide measured customer outcomes for the new integration.