---
title: How do ServiceNow AI agents connect to external agents via MCP and A2A?
url: https://opencontroller.lyzr.com/blog/servicenow-mcp-and-a2a-support
published: 2026-11-17
updated: 2026-11-17
author: Lyzr
---

# How do ServiceNow AI agents connect to external agents via MCP and A2A?

**ServiceNow supports three connection types: an MCP Server that publishes ServiceNow capabilities as tools for outside AI clients, an MCP Client that lets AI Agent Studio agents call external MCP servers, and A2A connections where ServiceNow acts as client or server to other agents. ServiceNow has grouped these under AI Agent Fabric and, more recently, Action Fabric. Each direction needs its own controls: tool exposure and OAuth for inbound MCP, server approval for outbound MCP, and per-agent credentials for A2A.**

Once ServiceNow agents can reach external tools and other vendors' agents, the governance question shifts from "what can this agent do in ServiceNow?" to "what can it reach, and what can reach it?". This guide covers each direction and the controls ServiceNow documents for it.

## What are the three connection types?

| Direction | What it does | Authentication (per ServiceNow) | Key control |
|---|---|---|---|
| MCP Server (inbound) | Publishes ServiceNow capabilities as MCP tools for external clients such as coding assistants or other agent platforms | OAuth | Which tools are exposed to which clients |
| MCP Client (outbound) | AI Agent Studio agents call external MCP servers | Connection and Credential alias in ServiceNow | Only approved servers are visible to agent builders |
| A2A | ServiceNow agents call, or are called by, agents on other platforms | OAuth 2.0 or API key, configured per external agent in AI Agent Studio | Which external agents are registered and with what credentials |

ServiceNow's explainer lists the tool types the MCP Server can publish: Now Assist Skills, Knowledge Graph, Subflows and Actions, and Scripted REST APIs. A Quickstart MCP Server ships with incident and case summarization tools. Minimum versions cited are Zurich Patch 9 or Australia Patch 2 for the MCP Server Console, and AI Agent Studio on Zurich or later for the MCP Client. ServiceNow has said it plans to move to A2A v1.0, so check the release notes for the protocol version your instance supports.

## How do you govern the inbound MCP Server?

Treat it like a public API with an AI caller.

1. **Start read-only.** Publish summarization and lookup tools first. Add write tools, such as Subflows or Scripted REST APIs that change records, only with a named owner and a use case.
2. **Scope per client.** Use the console's governance function to expose specific tools to specific clients, not the whole catalogue to every client.
3. **Rely on ACLs, then test them.** ServiceNow ACLs apply at the data layer. Test with a low-privilege user that a tool cannot return records the user should not see.
4. **Use lifecycle states.** Connections move through In Review, Assess, Approved, Active and Deprecated. Keep approval with the AI Steward role ServiceNow defines.
5. **Watch the audit trail.** AI Gateway reports usage, latency, success rate and errors per tool and per server. Alert on new clients and on spikes in write tools.

## How do you govern outbound MCP and A2A?

Outbound connections carry ServiceNow data to someone else's system.

- **Approve servers before builders see them.** Agent builders in AI Agent Studio see only approved MCP servers. Keep that gate and record why each server was approved. Our guide to [reviewing MCP servers agents already use](/blog/how-to-review-mcp-servers-agents-already-use) has a review template.
- **Prefer OAuth to API keys for A2A.** A shared API key hides which ServiceNow agent made the call. If you must use keys, scope one key per external agent and rotate it.
- **Verify the other agent.** For A2A, check the remote agent card and endpoint ownership before registering it. See [verifying A2A agent cards](/blog/verify-a2a-agent-cards).
- **Know what AI Gateway does not cover.** ServiceNow notes that AI Gateway governs MCP traffic, not A2A. You need logging and policy for A2A calls from another source.
- **Use the kill switch.** Admins can pause any client or server from AI Control Tower globally without redeploying. Rehearse it once so the team knows who can press it.

## What evidence should you keep?

For each connection, keep: owner, direction, protocol, authentication method, tools or skills exposed, data classes reachable, approval record, and the date of last review. That record supports access reviews and maps to agent risks in the [OWASP Top 10 for Agentic Applications](https://genai.owasp.org/), such as tool misuse and identity abuse. For the wider protocol choice, see [MCP vs A2A](/blog/mcp-vs-a2a-difference) and [how to govern A2A calls](/blog/how-to-govern-agent-to-agent-a2a-calls).

## How Opencontroller does this

Opencontroller keeps a governed registry of agents, versions, tools and MCP servers across platforms, and includes an Agent Gateway and an LLM Gateway for traffic you route through it. It supports open standards including MCP, OpenTelemetry, OIDC/SAML/SCIM and REST/webhooks, and its policy as code has built-in mappings to OWASP, MITRE ATLAS, NIST and ISO 42001. See the [MCP governance guide](/blog/how-to-govern-mcp-servers), the [cross-platform governance guide](/blog/how-to-govern-ai-agents-across-platforms), the [AI agent control plane overview](/blog/what-is-an-ai-agent-control-plane), or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Can external AI assistants call ServiceNow through MCP?**
Yes. The ServiceNow MCP Server publishes selected capabilities, such as Now Assist Skills and Subflows, as MCP tools that authorised external clients can call over OAuth.

**Does ServiceNow support the A2A protocol?**
Yes. ServiceNow agents can act as A2A client or server, with OAuth 2.0 or API key authentication configured per external agent in AI Agent Studio.

**Does ServiceNow AI Gateway govern A2A traffic?**
No. ServiceNow describes AI Gateway as governing MCP traffic only, so A2A calls need separate logging and policy.

## Key facts

ServiceNow Action Fabric (earlier AI Agent Fabric): MCP Server publishes Now Assist Skills, Knowledge Graph, Subflows and Actions, Scripted REST APIs as MCP tools (OAuth); Quickstart MCP Server has incident/case summarization; MCP Server Console minimum Zurich Patch 9 or Australia Patch 2. MCP Client in AI Agent Studio (Zurich+) uses Connection and Credential alias; builders see only approved servers. A2A: ServiceNow as client or server, OAuth 2.0 or API key per external agent; plans to update to A2A v1.0. Lifecycle states In Review, Assess, Approved, Active, Deprecated; AI Steward approves; pause clients/servers globally from AI Control Tower; AI Gateway audit per tool and server; AI Gateway governs MCP only, not A2A. Opencontroller: registry of agents, tools, MCP servers; Agent Gateway; LLM Gateway; open standards MCP, OTel, OIDC/SAML/SCIM. Pricing starts from $250K per deployment per year.

## Related

- [How do you govern ServiceNow AI agents?](https://opencontroller.lyzr.com/blog/govern-servicenow-ai-agents.md)
- [How do you govern agent-to-agent (A2A) calls?](https://opencontroller.lyzr.com/blog/how-to-govern-agent-to-agent-a2a-calls.md)
- [How do you verify an A2A agent card before trusting a remote agent?](https://opencontroller.lyzr.com/blog/verify-a2a-agent-cards.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
