Claude and ServiceNow Integration | MCP Connection Guide and ITSM Automation

Why Claude × ServiceNow Now — Background of the Anthropic Partnership

ServiceNow and Anthropic announced a partnership on January 28, 2026 to integrate Claude models into ServiceNow's core workflows, targeting the embedding of Claude into mission-critical operations including app development, healthcare, and life sciencesSource.

A defining milestone is that Claude was adopted as the standard model for ServiceNow Build Agent. This enables even non-specialist staff to build and deploy workflows using natural language. ServiceNow itself publicly stated that it deployed Claude to over 29,000 employees, reducing sales preparation effort by up to 95% and introducing Claude Code for engineering workSource.

In other words, "connecting Claude to ServiceNow" is no longer a niche experiment — it has become a vendor-approved primary use case. This article walks through the concrete steps to implement that connection in your own environment.

Three Ways to Connect Claude to ServiceNow

As of June 2026, there are three main approaches to connecting Claude and ServiceNow.

1. ServiceNow Native MCP (Zurich release): The path where ServiceNow officially supports both MCP server and client functions, governed by AI Control Tower. Best for organizations with strict enterprise governance requirements.
2. Open-source mcp-server-servicenow: A lightweight, community-built MCP server. Launchable via uvx, it provides the fastest way to get started for personal testing or proof-of-concept work.
3. Third-party managed MCP (Strac / Merge / Composio / CData, etc.): Hosted solutions that package OAuth, DLP, and auditing together. No code required; connection takes minutes to 10 minutes, with a governance layer includedSource.

From Claude's perspective, all three look like the same "MCP server." A practical approach is to quickly prototype with option ② and then move to option ① or ③ for production with proper governance in place.

What Is ServiceNow Native MCP (Zurich)?

With the Zurich release, ServiceNow added native MCP support, making it both a "consumer" (calling external tools) and a "provider" (opening its own capabilities to external AI). The "MCP Server Console" management UI allows you to control which tables and skills are exposed to external AISource.

Crucially, all operations from external AI go through AI Control Tower (AICT). The entire "system of action" — flows, playbooks, approvals, and catalogs — is executed with verified identity, scoped permissions, and full auditing. Rather than giving AI unconstrained database access, this design has AI work within ServiceNow's role model and governance framework.

Connection Steps for Claude Code

Below are the steps to register the open-source mcp-server-servicenow with Claude Code — the fastest way to get up and running.

1. Install uv (the Python toolchain that includes uvx) locally.
2. Enable API access on the ServiceNow side and prepare the instance URL, username, and password (or token) for the connection user.
3. In Claude Code, run claude mcp add servicenow -- uvx mcp-server-servicenow to register the MCP serverSource.
4. Set credentials (instance URL and authentication details) as environment variables so the server can reference the target instance.
5. Run /mcp in Claude Code and confirm that servicenow shows as connected.

When using managed MCP, step ③ above takes the form claude mcp add --transport http <name> <endpoint URL>, and on first use you complete authorization by logging into ServiceNow via a magic link or OAuthSource. Claude Desktop can also connect the same way by adding the URL and token to the mcpServers block in its configuration.

ITSM Operations Claude Can Perform After Integration

Once connected, Claude can perform the following ServiceNow operations from natural language promptsSource:

• Incident management: Query and update incidents filtered by priority, state, or assignment group; add work notes.
• Change request reference: Read change requests including schedule and approver information.
• Problem records: View problems and their linked incidents.
• Knowledge search: Search knowledge articles matching incident symptoms and cross-reference them.
• CMDB lookup: Retrieve configuration items (CIs) along with their relationships and infrastructure topology.
• Write actions: Create and update records, reassign ownership, and advance approvals (on behalf of authorized users).

For example, a sequence of tasks like "assign unassigned P1 incidents to the responsible team and attach relevant knowledge articles to the work notes" can be completed entirely through chat, without navigating between screens. The greatest value is compressing routine first-response work.

Security Considerations Before Deployment

Behind the convenience, the confidentiality of ServiceNow data demands the utmost care. Ticket bodies frequently contain employee and customer PII, pasted credentials, and secrets, and the CMDB can expose the entire infrastructure topology in a single querySource. Having this data flow directly into AI context is a significant risk.

In practice, make sure to design in the following three safeguards:

1. DLP-based masking: Insert a governance layer that detects and redacts sensitive data before it reaches the AI's context.
2. Approval gates for high-risk write operations: Restrict destructive operations such as record creation, updates, and approval advancement to go through an approval flow.
3. Audit log preservation: Record who did what to which record and when, retaining it as a compliance trail.

When choosing native MCP, AI Control Tower (AICT) handles most of these requirements out of the box. The safe migration path is: use open-source for rapid PoC, then move to a governed path for production.

Summary — Start with PoC, Prioritize Governance for Production

The integration of Claude and ServiceNow has become a "vendor-approved primary use case" thanks to the Anthropic partnership and native MCP in the Zurich release. The fastest path is connecting via the open-source claude mcp add servicenow -- uvx mcp-server-servicenow, which is enough to try out everything from incident queries to CMDB retrieval through chat.

However, in production, the risk of exposing PII, CMDB data, and credentials becomes a real concern. Move forward with a design that applies DLP, approval gates, and audit logging — or choose native/managed MCP with AI Control Tower — and delegate ITSM operations to AI within a controlled scope. The recommended starting point is simply having Claude execute a single incident query in a test environment.