Latest in AI
OpenAI and ServiceNow: ChatGPT MCP Plugin Availability and the 2027 Roadmap
The expanded partnership brings ServiceNow actions into ChatGPT, while several of the larger development, security and voice capabilities are still planned.
OpenAI and ServiceNow expanded their partnership on October 28, with a ServiceNow MCP plugin available in the OpenAI marketplace. Several larger development and security capabilities remain planned.
That distinction matters if you are deciding what to evaluate now rather than what to put into next year's architecture plan.
What is available now, and what comes later
The official announcement separates the rollout into several stages:
| Capability | Announced timing |
|---|---|
| ServiceNow MCP plugin in the OpenAI marketplace | Available October 28 |
| Enhanced plugin capabilities | Expected by the end of 2026 |
| Voice in Super Intelligence Control Tower | Expected Q1 2027 |
| OpenAI-powered security capabilities | Early access expected January 2027; GA planned Q2 |
| Codex in AI Workflow Factory | Expected in 2027 |
Interactive ChatGPT experiences remain previews. A complete pricing, plan and region matrix is not provided.
How ServiceNow actions fit into ChatGPT
Customers choose which capabilities their connected ServiceNow instance exposes. OpenAI says Action Fabric applies existing permissions and business rules to actions such as record updates.
The relevant technical foundation predates this announcement. ServiceNow's Action Fabric documentation describes domain-specific MCP servers for areas including IT service management, operations and the configuration management database. Its console provides a place to expose and manage those capabilities for external agents.
Model Context Protocol, or MCP, gives a client a structured way to discover and call tools. In this arrangement, ChatGPT is another interface to selected ServiceNow capabilities. It does not need to become the system that owns every workflow rule.
The important boundary is the action being exposed. Reading an incident, changing its assignment and closing it have different consequences, even when all three start with a sentence in the same chat window.
What changed from the existing integration approach
ServiceNow had already made governed enterprise actions available to external agents through its MCP server. Its May announcement explains that records are connected to workflows, approval chains and business rules.
The October expansion adds a more direct ChatGPT distribution surface and a broader OpenAI product roadmap. It should not be described as ServiceNow inventing MCP connectivity for the first time.
For an enterprise team, that is a useful way to evaluate the launch. If the underlying workflow is already approved, a new interface may make it easier to reach. If the workflow is poorly scoped, adding a conversational entry point does not resolve the missing controls.
What to check before a pilot
Start with a narrowly defined task and a test account whose permissions match the intended user. Check which records it can read, which actions it can invoke, and what happens when a request crosses an approval boundary.
Then inspect the record of the action. A conversational answer saying a ticket was updated should be checked against the actual ServiceNow state. The same principle applies to failed, partially completed and repeated requests.
These are implementation checks, not evidence that the new plugin has a specific defect. They matter because the value of the integration depends on reliable execution inside an existing operational system.
The useful outcome is fewer handoffs between conversation and workflow without losing the controls that made the workflow safe to run. The next milestone to watch is how much of that can be demonstrated with the available plugin, before the larger 2027 promises arrive.
I'm Paras. I work at the intersection of AI, engineering and distribution.
I help enterprise AI teams explain complex integrations clearly, connecting technical details with the decisions customers need to make.