Does the ServiceNow MCP server work with Claude?

Yes. A ServiceNow MCP server works with any MCP client, including Claude Cowork, Claude Code, and Claude Desktop, as well as Cursor and other AI agents. You point it at your instance, authenticate, and Claude can then query and act on records in plain language.

Is there an official ServiceNow MCP server?

Sort of, but not the kind you download. ServiceNow's Zurich-era platform support makes the instance itself an MCP client (in AI Agent Studio) and an MCP provider (via the MCP Server Console). To connect Claude to ServiceNow today, most teams use a community server such as echelon-ai-labs/servicenow-mcp.

Is the ServiceNow MCP server free?

The leading community servers are open source (MIT) and free to run. You still need a ServiceNow instance and credentials (basic auth, OAuth, or an API key) with appropriate roles, and write actions in production should go through change control.

ServiceNow MCP Server: What It Does & How to Use It with Claude

Q: What can the ServiceNow MCP server do?

Depending on the loaded tool package, it can create and update incidents, manage change requests and approvals, work the service catalog and knowledge base, query the CMDB, and handle users, groups, and scripting. Start narrow — read-only on incidents — before granting write access.

The ServiceNow MCP server lets an AI assistant work inside your ServiceNow instance. It can query and create incidents, manage change requests and approvals, work the service catalog and knowledge base, and read the CMDB. Connect it to Claude Cowork, Claude Code, or Claude Desktop and you can ask Claude to triage, summarize, and act on ServiceNow records in plain language.

What can the ServiceNow MCP server do?

Capabilities depend on which tool package you load (servers like echelon-ai-labs/servicenow-mcp ship role-based packages plus a full package). Across the leading community servers, the surface looks like this:

Capability	What it enables
Incidents	Create, update, comment on, and resolve incidents; search and list by caller, state, priority, or assignment group
Change management	Create change requests, manage change tasks, move workflow states, and route approvals
Service catalog	Manage catalog items, categories, and variables; surface optimization recommendations
Knowledge base	Create and manage knowledge articles, categories, and knowledge bases
CMDB & records	Query configuration items and arbitrary tables to pull context behind an incident or change
Users & platform	Manage users and groups; work with workflows, script includes, UI policies, and changesets/update sets

In practice that means Claude can answer “what P1 incidents are open for the payments team?”, draft a change request for your review, or pull the CMDB record behind a recurring outage, all without leaving the assistant.

Is it an official or a community server?

ServiceNow’s native MCP support (Zurich). The platform itself acts as an MCP client (add external servers in AI Agent Studio) and an MCP provider (expose ServiceNow capabilities as governed tools via the MCP Server Console, governed through AI Control Tower). This is the right path for ServiceNow-hosted agents and Now Assist, but it is not a server you drop into Claude Desktop. It requires Zurich (Patch 4+), a Now Assist SKU, and the sn_mcp_server plugin.
Community echelon-ai-labs/servicenow-mcp (MIT, open source). This is the practical way to connect Claude to ServiceNow today. It ships role-based tool packages (service_desk, catalog_builder, change_coordinator, knowledge_author, platform_developer, system_administrator, agile_management) plus a full package, and supports basic auth, OAuth, or API key. michaelbuckner/servicenow-mcp (also MIT) is another widely used option.

How do you connect it to Claude (Cowork / Code / Desktop)?

Pick a community server (e.g. echelon-ai-labs/servicenow-mcp) and create credentials on your instance (basic auth, OAuth client ID/secret, or an API key) tied to a role with the minimum access you need.
Register the MCP server with your client. In Claude Cowork and Claude Desktop you do this via MCP settings (passing instance URL and credentials as env vars), or in Claude Code via claude mcp add.
Load the narrowest tool package (start with service_desk, read-only) against a sub-prod instance, confirm it behaves, then widen scope.

How we use it: ServiceNow MCP is genuinely good for triage and reporting: “summarize this incident’s activity,” “which changes are pending CAB approval this week,” “draft a KB article from this resolved ticket.” It’s bad for blind bulk writes: never let an agent close, reassign, or mass-update records in prod without a human in the loop. The most common gotcha is testing against production and tripping ACLs or business rules you forgot existed. Always start on a dev/test instance with a least-privilege role.

Security & permissions

Treat ServiceNow like production data: scope the integration user to a least-privilege role, default to read-only, route write actions through normal change control, and prefer OAuth over shared basic-auth credentials for anything beyond a personal test instance.

Wiring ServiceNow into your stack

If you’d rather have ServiceNow (and the rest of your tools) wired into a Claude setup that fits your workflows, scoped safely, governed, and maintained, that’s what we do. Book a discovery call.

Sources: ServiceNow MCP client docs · Implementing MCP in ServiceNow · echelon-ai-labs/servicenow-mcp · michaelbuckner/servicenow-mcp.