SujanDutta
Administrator
Options
- Subscribe to RSS Feed
- Mark as New
- Mark as Read
- Bookmark
- Subscribe
- Printer Friendly Page
- Report Inappropriate Content
a month ago
All You Need to Know About ServiceNow Action Fabric
ServiceNow Action Fabric enables AI agents on the Now Platform to work with external agents and tools through MCP and A2A protocols. It supports three patterns: ServiceNow as MCP client (calling outward), ServiceNow as MCP server (exposing capabilities), and ServiceNow agents collaborating with external agents via Google's A2A protocol.
This series examines each Action Fabric concept weekly to help you get the most out of AI agents on the platform.
Table of Contents
- Episode 1: Action Fabric Intro
- Episode 2: MCP in ServiceNow
- Episode 3: ServiceNow as MCP Client
- Episode 4: ServiceNow as MCP Server
- Episode 5: A2A (Agent2Agent) in ServiceNow
- Frequently Asked Questions
Episode 1: Action Fabric Intro
The Problem: Automation Islands
Today's enterprises run multiple applications, each shipping its own AI agents. These agents create isolated automation silos where they cannot coordinate across platform boundaries. The result: you can't build modular workflows that span vendors.
The Solution: Cross-Platform Interoperability
ServiceNow Action Fabric delivers orchestration where agents across your technology stack interoperate, with ServiceNow at the center.
The Tech Stack Behind the Concept
Action Fabric is platform-agnostic, delivering interoperability through three protocols:
1. ServiceNow as MCP Client (Outbound)
ServiceNow AI agents initiate outbound calls to external MCP servers, discovering published tools and retrieving context or capabilities. This extends agent reach by connecting to specialized external tools.
2. ServiceNow as MCP Server (Inbound)
ServiceNow publishes its capabilities as MCP tools that external clients like ChatGPT or Cursor can consume. Now Assist skills, custom skills, scripted APIs, and knowledge graphs all become available as tools.
3. A2A (Agent2Agent)
Google's open Agent2Agent protocol enables two AI agents to coordinate as peers on shared tasks, delegating subtasks and passing context across turns. ServiceNow agents can act as primary agents delegating outward or secondary agents receiving external delegations.
Episode 2: MCP in ServiceNow
MCP as the USB-C Moment
Before Model Context Protocol, every agent required custom integration code to communicate with each tool or system.
MCP is the USB-C moment. One standard protocol. Any agent can connect to any MCP server, no custom integration work required.
ServiceNow as MCP Server
Existing ServiceNow capabilities become tools any external AI can call. Now Assist skills, scripted REST APIs, and Table API queries packaged behind an MCP server become accessible to any external LLM or agent speaking MCP. Discovery is simple: once agents discover available tools, they make tool calls and ServiceNow returns results.
ServiceNow as MCP Client
ServiceNow reaches outward as an MCP client, consuming tools from external servers such as Microsoft Learn MCP servers or other MCP-compliant sources.
MCP provides the infrastructure for enterprise automation flows.
Episode 3: ServiceNow as MCP Client
The Use Case: Resolution Assist Agent
Support users handling incidents often need information that lives outside the platform: Microsoft application documentation, vendor guidance, or external knowledge bases. MCP as client closes this gap by letting ServiceNow AI agents discover and consume external tools while staying inside the incident workflow.
Workflow Steps:
- Support user handles an incident
- Agent pulls incident context (short description, full description) using an internal script tool
- Agent reaches out to an external MCP server (Microsoft Learn in this demo), performing a doc search using that context
- Retrieved guidance posts to the incident as a work note for user review or knowledge article creation
This demonstrates the mechanics clearly enough to adapt to your own external MCP servers and workflows.
Technical Implementation
1. Register the External MCP Server
In AI Agent Studio, navigate to Settings > Manage MCP Servers. This landing page registers external MCP servers, enabling ServiceNow to discover their published tools.
- Choose authentication type: OAuth 2.0 or API key
- For API keys, provide the server URL and key
- Saved servers appear as available options
2. Create an AI Agent
Add a new AI agent with a clear, meaningful role description so it understands its function. Default settings work for initial attempts.
3. Add Internal Script Tool for Incident Context
Configure a simple script tool that fetches incident details (short description and description fields) using the incident number. Set it as supervised so the agent confirms before executing.
4. Add the MCP Server Tool
Add tool type "MCP tool" and select the registered MCP server. ServiceNow performs a tools/list discovery call against the server; every exposed tool appears as an option.
5. Add Script Tool to Post Results
Configure a second script tool that takes the incident number and the resolution content returned by the MCP tool, then writes the results back as an incident work note.
6. Test It
With the reasoning panel enabled:
- Agent pulls incident context and identifies the actual issue
- Supervised tool pauses for confirmation before proceeding
- Agent proposes a search query and requests confirmation
- Upon confirmation, agent invokes the MCP tool, reaching the external server and returning documentation-based guidance unavailable in the incident or internal knowledge base
- Agent automatically posts guidance to work notes
Result: Support users receive externally sourced, relevant resolution guidance directly in incidents without leaving ServiceNow.
Any agent that needs external context, data, or capability can reach through registered MCP servers under governed authentication, bringing results into triggered records or workflows.
Episode 4: ServiceNow as MCP Server
ServiceNow can expose skills, REST APIs, flows, sub-flows, actions, and knowledge graphs as tools for external MCP clients like Claude or ChatGPT.
The Use Case: ServiceNow as Tool Provider for External AI
This scenario keeps the mechanics visible:
- Register a new MCP server inside ServiceNow and attach tools (starting with incident lookup and modification)
- Set up a secure inbound integration path (OAuth, JWT tokens, user-scoped access) enabling external AI access
- Connect an external AI client (Claude in this demo) to the ServiceNow MCP server using a custom connector
- Extend the server by exposing a custom scripted REST API as a tool
- Test: create an incident from natural language, add a work note, pull data through a custom REST API, all from outside ServiceNow
Technical Implementation
1. Open the MCP Server Console
Requires AI admin role; instance needs Zurich Patch 9 or later, or Australia Patch 2 or later. The out-of-the-box "Quick Search" MCP server provides a sandbox for understanding MCP server behavior before building custom ones.
2. Create New MCP Server
Provide a meaningful name and short description. Upon saving, the server URL populates automatically; this is the endpoint external clients connect to. Prefer multiple MCP servers over one large one for governance. This keeps access boundaries clean.
3. Add Tools to Server
Start simply. In this demo, two tools get added: a lookup incident records tool and a modify incident tool.
4. Set Up Inbound Integration (Machine Identity Console)
This enables external agents to authenticate against the new MCP server.
- Open Machine Identity Console and set up an application registry entry
- Choose Authorization Code grant type
- Fill in the redirect URL
- Copy the generated client ID and client secret
- Scope broadly for demos; production requires locking down API scopes
- Switch token format to JWT (currently expected by the MCP server connection)
5. Build Connection from External Client (Claude)
- In Claude's connector settings, add a custom connector
- Name it (e.g., "ServiceNow MCP") and paste the MCP server URL, client ID, and client secret
- The connector registers but is not yet connected; authentication is still needed
- When connecting, you are prompted for ServiceNow credentials. This keeps the connection user-scoped, so the external AI only sees what that specific user account is entitled to view (no elevated privileges from inbound requests)
6. Attach Now Assist Skill to MCP Server
Create a custom analysis skill that accepts natural-language queries to create incidents. Create a new tool, mark it as "analysis skill" type, and attach the MCP server. Single tools can attach to multiple MCP servers for context reuse.
7. Expose Scripted REST API as Tool
Beyond Now Assist skills, you can expose existing scripted REST APIs, flows, sub-flows, actions, and knowledge graphs as MCP tools. In this demo, a simple scripted REST API tied to incidents gets registered as a tool, tagged to the same MCP server, given a description, and saved.
8. Refresh Tool List on Connector
Back on the Claude side, refresh the MCP server's tool list. Newly added tools appear automatically; no reconnect needed, just refresh.
9. Test End-to-End
- Create incident from natural language: "My MacBook is extremely slow" + "it's really high priority." Claude calls the MCP server, discovers available tools, matches the request to the incident-creation tool, and creates the incident.
- Add work note: "Can you add a comment to this incident?" Claude identifies the modify incident tool and adds the comment.
- Pull data through custom REST API tool: Claude calls the scripted REST API tool and returns the expected results, confirming the custom-exposed tool works like the built-in ones.
Result: External AI clients operating under the logged-in user's permissions create and update ServiceNow records and call custom APIs without opening the ServiceNow UI.
Action Fabric's MCP server direction makes your existing platform investment (skills, APIs, knowledge graphs) into reusable infrastructure for whatever AI tooling your organization or users prefer.
Episode 5: A2A (Agent2Agent) in ServiceNow
By this point in the series, Action Fabric and MCP are old news: you know ServiceNow can act as an MCP client, reaching out to external tools, and as an MCP server, exposing its own capabilities to the outside world. The next question, the one Sayan tackles this episode, is what happens when a tool call isn't enough and you actually need one AI agent talking to another.
What A2A Actually Is
A2A is open standard protocol for direct, peer-to-peer collaboration between AI agents. It's a different shape than MCP. MCP is an agent talking to a tool: a request, a response, like calling a function. A2A is an agent talking to another agent: it hands over structured context, the other agent goes and does its own work, and it reports back. That exchange doesn't have to end after one round trip. It can run as a real multi-turn conversation, back and forth, until the job is actually finished.
A Job for MCP
Picture Claude sending a query into ServiceNow and asking for a list of knowledge articles back. Build the skill, wrap it as a tool, attach it to the server: one call in, one clean result out. That's a textbook MCP job: a single function that returns a single answer.
A Job for A2A
Now picture asking an agent that lives somewhere else entirely (built on Google ADK, LangGraph, whatever framework) to plan a trip. That agent isn't going to hand you a flight itinerary on the first try. It comes back with questions: where are you leaving from, where are you headed, when, flight or train? You answer, it goes and looks at options, and the back-and-forth keeps going until you've landed on something. That multi-turn, keep-talking-until-it's-solved pattern is where A2A earns its keep.
How ServiceNow Plays Both Sides
ServiceNow already knows how to be an MCP client and an MCP server. A2A adds a second pattern on top of that same foundation, and ServiceNow can sit on either side of it.
ServiceNow as the primary agent. Here, ServiceNow's AI agent is the orchestrator running the workflow, and it reaches out to an external agent: a Google agent, an AWS agent, an agent built on any framework. The sequence is straightforward: discover the agent, pull everything it can do, then execute it and delegate the task. Once that happens, ServiceNow is the primary agent and whatever it called is the secondary agent.
ServiceNow as the secondary agent. Flip it around, and ServiceNow can just as easily be the one getting called. Register an AI agent in Agent Studio, make it discoverable, and external third-party agents can find it, run it, and delegate work to it: the same interoperability, just running in the other direction.
That's the core of A2A: MCP still owns the clean, single function call. A2A is what you reach for the moment the job means handing real work to another agent and following it through to the end.
Frequently Asked Questions
What is ServiceNow Action Fabric used for?
ServiceNow Action Fabric makes AI agents interoperable across vendor boundaries, enabling agents built on ServiceNow to work with agents and tools from Microsoft, SAP, Salesforce, and other platforms. It provides cross-platform agent orchestration with ServiceNow at the network center. Practical outcome: single workflows span multiple vendors' agents without custom point-to-point integration.
What is the difference between Action Fabric and MCP?
MCP (Model Context Protocol) is an open standard for exposing and consuming tools between AI systems. Action Fabric is the ServiceNow framework that implements MCP alongside Google's A2A protocol as a governed capability on the Now Platform. MCP is the protocol; Action Fabric is how ServiceNow makes that protocol usable with platform authentication, scoping, and agent orchestration.
What is the difference between MCP and A2A in ServiceNow?
MCP operates at the tool level: it lets AI agents discover and call individual capabilities like Now Assist skills, scripted REST APIs, or knowledge graph lookups. A2A (Agent2Agent, Google's open protocol) operates at the agent level, letting ServiceNow AI agents delegate to and collaborate with complete external agents. In Action Fabric, MCP covers tool access in both directions while A2A covers peer-to-peer agent interaction.
Can ServiceNow agents talk to ChatGPT or Cursor?
Yes, in both directions. When ServiceNow acts as MCP server, external clients like ChatGPT and Cursor discover and call Now Assist skills, scripted REST APIs, and knowledge graphs, subject to OAuth and user-scope enforcement (the caller only sees what that identity is entitled to). When ServiceNow acts as MCP client, agents registered in AI Agent Studio reach outward and call tools on any registered external MCP server.
Is Action Fabric part of Now Assist?
No. Action Fabric and Now Assist are separate but complementary. Now Assist provides generative AI skills running on the platform; Action Fabric is the interoperability layer that lets those skills be exposed to external AI systems and lets ServiceNow agents consume external tools. In practice, Now Assist skills become tools published through Action Fabric MCP servers.
What does it mean for ServiceNow to act as an MCP client?
ServiceNow AI agents initiate an outbound connection to an external MCP server, discover published tools, and call them to retrieve information or capability that does not exist inside ServiceNow. The agent stays in control of the workflow; the external server supplies the tool call and result.
What authentication does ServiceNow support when registering an external MCP server?
The Manage MCP Servers page supports OAuth 2.0 and API key authentication. For API key auth, provide the server URL and key directly when registering the server.
What's the difference between ServiceNow as an MCP client vs. an MCP server?
As MCP client, ServiceNow initiates an outbound call pulling external tools or data. As MCP server, ServiceNow is on the receiving end, with external AI clients connecting and calling exposed tools like Now Assist skills, scripted REST APIs, and knowledge graphs.
What can I expose as an MCP tool besides Now Assist skills?
Scripted REST APIs, flows, sub-flows, actions, and knowledge graphs can all register as MCP tools alongside table lookups like incident records.
Does the external AI need to know which tool to call?
No. That is the point of MCP discovery. The client calls the MCP server, retrieves the available tools list, and matches the natural-language request to the right tool automatically. For example, it recognizes "add a comment" should call the modify-incident tool, not the incident-creation tool.
Can I reuse the same tool across multiple MCP servers?
Yes. Single tools can attach to multiple MCP servers if relevant to multiple use cases.
What is A2A in ServiceNow?
A2A (Agent2Agent) is open protocol for direct, peer-to-peer collaboration between AI agents. Inside Action Fabric, it lets a ServiceNow AI agent hand off a task to an external agent, or receive one from an external agent, using structured context instead of a single tool call.
How is A2A different from MCP?
MCP is an agent calling a tool: one request, one response, similar to calling a function. A2A is an agent calling another agent: it passes structured context, the other agent runs its own process, and the two can go back and forth across multiple turns before the task is done.
When should I reach for MCP instead of A2A?
If a single function call gets you what you need, MCP does the job: a query in, one clean result out, like ServiceNow returning a list of knowledge articles to Claude.
When should I reach for A2A instead of MCP?
Whenever the task needs real back-and-forth with another agent rather than one call and done, for example delegating a multi-step job like trip planning to an external agent that asks follow-up questions before it can complete the work.
What does it mean for ServiceNow to be the "primary agent" in A2A?
ServiceNow is the primary agent when its AI agent acts as the orchestrator: it discovers an external agent, retrieves that agent's capabilities, then executes and delegates the task to it.
What does it mean for ServiceNow to be the "secondary agent" in A2A?
ServiceNow is the secondary agent when an external agent is the one reaching in. Once an AI agent is registered in Agent Studio and enabled for discovery, outside agents can find it, execute it, and delegate work to it.
Can ServiceNow be both the primary and secondary agent?
Yes. The two patterns aren't exclusive: the same ServiceNow instance can call out to external agents as a primary agent and be discoverable by external agents as a secondary agent.
Does A2A support multi-turn conversations between agents?
Yes, that's the point of it. An A2A exchange can run across multiple turns while the two agents work through a task together, instead of stopping after one call.
What kinds of external agents can ServiceNow talk to over A2A?
Any agent built to speak the A2A protocol, regardless of framework: Google ADK, LangGraph, AWS-based agents, or anything else that supports the standard.
What's Next
Next week, we look at ServiceNow AI agents collaborating with external agents over the A2A protocol.
- 2,934 Views
2 Comments
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
